ivoras' Home

FreeBSD under VMWare

The following tips work both under VMWare ESX (2 & 3) and VMWare Server 1.0.

1. General tips

Don't use a virtual machine for network-heavy workloads. VMware and other full-hardware virtualization environments (MS Virtual PC, QEmu, etc.) introduce a heavy penalty on I/O, especially network I/O. Expect to be able to get only around 30%-40% out of a gigabit interface (which still amounts to ~~40 MB/s). For example, don't use virtual machines for network routers and similar tasks. Also, don't use them for tasks which require exact timing (e.g. multimedia processing, industrial machine control, etc.). These points are actually valid for any combination of virtualization software and guest OS.

2. Don't use `lnc`

This tip is no longer current, as lnc doesn't exist in FreeBSD 7. It's still valid for FreeBSD 6.3 and earlier versions.

While it's the default, lnc driver is the worst network driver for your virtual machine. It's GIANT-locked (meaning it doesn't allow for much parallelizm in the OS), and it's actually deprecated and will be dropped in FreeBSD 7. The replacement for lnc is le and it's present at least in FreeBSD 6.2 and newer, but it's not included in the default GENERIC kernel. Thus, you'll have to configure and compile a custom kernel with device lnc replaced with device le. (Just loading the if_le kernel module won't work because the lnc driver present in the kernel at boot time will detect and use the hardware first.)

In VMWare ESX 3 and VMWare Server 1.0 there's an undocumented configuration option for the virtual machine that enables VMWare to emulate Intel E1000 hardware instead of the AMD Lance^[*]. To use it, edit the .vmx file and put ethernet0.virtualDev="e1000" in the appropriate place (anywhere). The simulated device also has TSO support (which is usable in FreeBSD 7, though I don't know what performance can be achieved with the simulated hardware). Device em is faster and not GIANT-locked so it should give the best performance.

3. Reduce `kern.hz`

Kernel's timer frequency ("HZ") in FreeBSD 6.x and 7.x is set to a relatively large value - 1000 Hz. This was done in attempt to reduce latency in the interim period while some kinks are sorted out and it's planed for it to be reduced back to 100 Hz in future versions of FreeBSD. High HZ setting has a negative impact on simulated machine's performance because the VM software spends a lot of time handling timer interrupts instead of doing real work.

You can change the HZ setting by adding a line like kern.hz=100 in /boot/loader.conf. In the extreme cases you might want to use very low values for kern.hz like 20, but test first!

4. Disable internal VMWare swapping

Consider disabling VMWare internal memory swapping and make the virtual machine fit in in the physical memory of the host for best performance.

5. VMWare Tools not necessary

It would be nice to have VMWare Tools working on FreeBSD but apparently the company doesn't want to support it properly. Currently, the only thing that VMWare tools are useful on FreeBSD is to get GUI features like clipboard sharing and automatic mouse focus grab in X.Org. VMWare tools on Linux seem to include a driver that does something with memory management, but it's not available for FreeBSD. You don't need VMWare Tools for the following things to work: networking, timer, X.Org GUI.

Networking is handled by the le driver or the em driver. These two will work without any special configuration of FreeBSD. To use the em driver, you might need to modify the VM configuration to include ethernet0.virtualDev = "e1000" or a similar appropriate line. To use the VMWare vmxnet driver (which as far as I can see isn't much different than the le driver), you need to build a kernel without the le driver first.

Timer issues can be lessened (never resolved, even with VMware tools) by reducing kern.hz to something like 50 or 100 Hz (in loader.conf), and installing ntpd.

X.Org can use the generic "vmware" display driver which is included in the default X.Org collection of drivers. Mouse, etc. are also handled generically.

The only remaining functionality is the ability to "shrink" drives. If you're running VMWare Server on Windows host, there are third party utilities for VMWare that can do the same thing.

^{[*]If you're from VMWare-the-company,
please don't remove this feature from both of the products.
It's very useful both in ESX and Server for both production and
testing.}

What's cooking for FreeBSD 8?

The next major release of FreeBSD, version 8, will probably be an "evolutional" release. The focus will be on further improvements of SMP capabilities and the stability of already implemented subsystems. This page will document changes that will be included in FreeBSD 8, including those that might end up being committed to earlier branches. In other words, it describes differences between 7.0 and 8.0, no matter what happens to the versions in between.

Also useful are the quarterly Status Reports:

Everyone's encouraged to download a snapshot CD image and try all the new features (as well as the old ones). Developers are very interested in bug reports. Note that FreeBSD 8.0 is not released yet and both the snapshots and the default source trees have debugging enabled by default (which results in dramatic slowdowns so don't benchmark them without removing the debugging options).

If you're interested in how FreeBSD gets developed, you're encouraged to read the mailing lists and developer blogs.

Kernel & low level improvements

MPSAFETTY

Status: In final stages of development
Will appear in 8.0: probably
Author: Ed Schouten
Web: wiki page

The TTY layer is the traditional Unix interface to system users, providing them with interactive sessions to run shells, etc. The current TTY layer in FreeBSD is for the most part the traditional BSD TTY, which is integrated with the drivers and other layers in a way that, though efficient, makes it hard to maintain and extend. The initiative to rewrite the TTY layer aims to make it a true abstraction layer, operating on behalf of both sides of TTY. In addition, it will remove the TTY from the Giant lock, which will eliminate problems with lags and skippy user interface behaviour in the console and X.Org.

Kernel memory limit on AMD64 increased to 6 GB

Status: Committed to -CURRENT
Will appear in 8.0: sure
Author: Alan Cox
Web: announcement

Some modern features (of which the most notable currently is ZFS) require a large amount of kernel memory (this has nothing to do with traditional disk caches). Up to now, it was only possible to allocate up to 2 GB for kmem_max, which is becoming a bit cramped. There is ongoing work to increase this to at least 6 GB. Users of ZFS will be happy.

Kernel threads

Status: Committed to -CURRENT
Will appear in 8.0: sure
Author: Julian Elischer
Web: commit message

Kernel threads upto now were actually "heavy weight" processes running in the kernel address space. This change introduces real light weight kernel threads which consume less low-level resources (process locks, memory maps). It also allows better grouping of threads for display purposes.

procstat(1): A process inspection utility

Status: Committed to -CURRENT
Will appear in 8.0: sure
Author: Robert Watson
Web: announcement

procstat combines functionality from the now-deprecated procfs(4) and adds several new functionalities. Some of the data procstat can provide are: process' command line arguments, file descriptor information, stacks of the kernel threads in the process, security credentials information from the process, thread information and virtual memory mappings. This is utility is mostly useful for debugging.

TextDumps: gathering information after kernel panic

Status: Committed to -CURRENT, MFC-ed to 7-STABLE
Will appear in 8.0: sure
Author: Robert Watson
Web: Q&A on textdumps

The usual thing that happens after a kernel panic is a kernel memory dump, either full or (in 7.0 and later) a "minidump". The new "textdump" feature doesn't store the actual kernel memory dump, but extracts commonly needed information from it, stores it into a tar archive of text files, and deletes the dump file. This significantly reduces the size requirements of collecting such information, speeds up development, and enables people to collect debugging information after a crash without kernel developer experience.

ULE 3.0: New version of the SMP-optimized scheduler

Status: Committed to -CURRENT
Will appear in 8.0: sure
Author: Jeff Roberson
Web: commit message, commit message, announcement

Evolution of the ULE scheduler resulted in support for fine-grained CPU affinity calculations, taking into account the physical topology of the CPUs (caches, cores, sockets) and much improved support for binding threads to CPUs. This results in additional functionalities (opens up the possibility of assigning individual CPUs to jails) and noticeable performance improvements.

Superpages

Status: Committed to -CURRENT
Will appear in 8.0: sure
Author Alan Coy
Web: research paper

Most general-purpose processors provide support for memory pages of large sizes, called superpages. Superpages enable each entry in the translation lookaside buffer (TLB) to map a large physical memory region into a virtual address space. This dramatically increases TLB coverage, reduces TLB misses, and promises performance improvements for many applications. However, supporting superpages poses several challenges to the operating system, in terms of superpage allocation and promotion tradeoffs, fragmentation control, etc. We analyze these issues, and propose the design of an effective superpage management system. We implement it in FreeBSD on the Alpha CPU, and evaluate it on real workloads and benchmarks. We obtain substantial performance benefits, often exceeding 30%; these benefits are sustained even under stressful workload scenarios.

For now, superpages are only available for AMD64 architecture (aka x86_64).

DTrace

Status: Committed to -CURRENT (kernel trace only)
Will appear in 8.0: sure
Author: John Birrell
Web: project web page

DTrace is a tool and a language developed by Sun Microsystems to help debugging and profiling operating systems. It can aggregate information from different parts of both kernel and userland and analyze them in a ways that's meaningful to the user.

Networking improvements

VirtNet / VIMAGE / Imunes / Network stack virtualization

Status: In final stages of development
Will appear in 8.0: sure
Author: Marko Zec
Web: project web page

The network stack virtualization project aims at extending the FreeBSD kernel to maintain multiple independent instances of networking state. This will allow for complete networking independence between jails on a system, including giving each jail its own firewall, virtual network interfaces, rate limiting, routing tables, and IPSEC configuration.

Multiple routing tables / FIBs

Status: Committed to 8-CURRENT
Will appear in 8.0: sure
Authors: Julian Elischer
Web: proposal

Support for multiple routing tables (forwarding information bases) allows advanced network topologies. The setfib utility can be used to select routing tables on per-process level.

Equal cost multipath routing

Status: Committed to 8-CURRENT
Will appear in 8.0: sure
Authors: Qing Li
Web: commit message

ECMP routing allows for multiple routes to be handled by the kernel, including default routes. It potentially offers substantial increases in bandwidth by load-balancing traffic over multiple paths.

Zero-copy BPF

Status: Committed to 8-CURRENT
Will appear in 8.0: sure
Authors: Robert Watson, Christian S.J. Peron
Web: BSDCan slides

BPF is Berkeley Packet Filter, facility used to capture raw network packets from the lower layers of the network stack according to user-defined filters and forward them to an application, as well as insert raw packets to the network stack.

This improvement to BPF reduces the number of memory copy operations between the kernel and the application which improves performance in some cases.

Kernel NFS locking support

Status: Committed to 8-CURRENT
Will appear in 8.0: sure
Author: Doug Rabson
Web: commit message, announcement

NFS lock manager in kernel improves performance and behaviour of NFS locking (used to synchronize file access on remote machines). New features include multithreaded operation, deadlock detection, and transparent interaction with local file locks on the server.

Storage subsystems' improvements

Boot support for GPT partitions

Status: Committed to -CURRENT
Will appear in 8.0: sure
Author: John Baldwin
Web: commit message

Support for booting from GPT partitions has been committed to -CURRENT. This support includes the boot sector and loader that enable common i386 machines with a generic BIOS to boot from GPT-partitioned drives.

bsdlabel gets extended to 26 partitions

Status: Committed to -CURRENT
Will appear in 8.0: sure
Author: Marcel Moolenaar
Web: commit message

bsdlabel is (finally!) extended to support more than 8 partitions. The new limit of 26 partitions comes from the number of lower-case leters.

To make use of this change, GEOM_PART needs to be used instead of GEOM_BSD (this requires custom kernel configurations).

Other changes

User-controllable CPU/IRQ binding (jhb)
User-controllable CPU-thread binding with support for CPU sets (jeffr)

My CV

Name: Ivan Voras
Title: dipl.ing. (M.Sc. CSE)
Academia: PhD grad student
Currently employed: University of Zagreb, Croatia, Faculty of electrical engineering and computing (Expert associate, Internet services architect)

My CV is available on request. I'm usually available for temporary contracting work in the areas of designing and deploying Unix systems, system-level programming in C & similar jobs, with some experience in development team leadership.

My FreeBSD things

I've come to have so many things related to the FreeBSD project that I finally went and organised all into one section on the web. On the downside, the font of the sections' listing is getting smaller and smaller :)

This page lists only the smaller projects, patches, etc. - bigger projects have their own separate pages. See here for more projects.

What's cooking for FreeBSD 8?

See this page if you're interested.

There's also my blog with FreeBSD-related themes.

Miscellaneous

Short tutorial on writing FreeBSD GEOM classes, and on FreeBSD kernel programming in general.

Encrypted iSCSI device tutorial

Suppose you want to use a remote iSCSI device, but you don't exactly trust either the storage or the network in between. Of course, there's a way around it :)

The setup presented here is very simple and will behave like this:

[iSCSI server] -- encrypted data on the server and over the wire -- [iSCSI client]

Note: all these instructions are valid for FreeBSD 7.0 - previous versions are probably missing some parts.

Setting up an iSCSI target

You can skip this section if you already have an iSCSI target (a "target" is where the data is stored, i.e. the "server" node of iSCSI).

1. Install the iscsi-target port.

2. Edit /usr/local/etc/iscsi/targets file and add lines similar to the following:

# NAME          DEVICE          START           LENGTH
extent0         /dev/da2        0               5GB
# NAME          ACCESS          STORAGE         NETMASK
target0         rw              extent0         10.0.0.0/24

These lines should be self-explanatory. If you need more help, see targets(5) or NetBSD's iscsi-target HOWTO.

3. Enable iscsi-target in /etc/rc.conf by adding the following line to it:

iscsi_target_enable="YES"

4. Start the server by running /usr/local/etc/rc.d/iscsi_target start. You should see something like the following outputted to the console:

Starting iscsi_target.
Reading configuration from `/usr/local/etc/iscsi/targets'
target0:rw:10.0.0.0/24
        extent0:/dev/da2:0:5368709120
DISK: 1 logical unit (10485760 blocks, 512 bytes/block), type iscsi fs
DISK: LUN 0: 5120 MB disk storage for "target0"
TARGET: TargetName is iqn.1994-04.org.netbsd.iscsi-target

Setting up the iSCSI initiator

The "initiator" is the client part in iSCSI, and it connects to the server. The following steps should be done on the client system.

1. Edit /etc/iscsi.conf and add the following lines:

target0 { # nickname
        targetaddress        = 10.0.0.102
        targetname           = iqn.1994-04.org.netbsd.iscsi-target:target0
}

2. Load the iscsi_initiator kernel module with:

# kldload iscsi_initiator

Also, add the following line to /etc/loader.conf to load the module on boot:

iscsi_initiator_load="YES"

3. Start the iSCSI session by running:

# iscontrol -n target0

Several lines should be output to the console, which should look like the following:

iscontrol[8516]: running
iscontrol[8516]: (pass3:iscsi0:0:0:0):  tagged openings now 0
iscontrol[8516]: cam_open_btl: no passthrough device found at 1:0:1
iscontrol[8516]: cam_open_btl: no passthrough device found at 1:0:2
iscontrol[8516]: cam_open_btl: no passthrough device found at 1:0:3
iscontrol: supervise starting main loop

More importantly, the kernel log (which you can see with tail /var/log/messages) should now contain something similar to this output:

Jan  4 23:17:08 client kernel: da0 at iscsi0 bus 0 target 0 lun 0
Jan  4 23:17:08 client kernel: da0:  Fixed Direct Access SCSI-3 device

This means the device da0 has been created - this is the local representation of the remote iSCSI drive.

3. Set up GEOM_GELI on the new device:

# geli init /dev/da0

The utility will ask you for a passphrase which will be used to encrypt the data. GEOM_ELI (as is the encryption layer known) has many more options, but the defaults are good enough. It will use AES encryption with sane defaults.

4. Load the GEOM_ELI kernel module:

# kldload geom_eli.ko

Also, add the following to /boot/loader.conf to load the module at boot time:

geom_eli_load="YES"

5. Attach the encrypted device:

# geli attach /dev/da0

Lines similar to the following should appear in the kernel log:

Jan  4 23:33:28 client kernel: GEOM_ELI: Device da0.eli created.
Jan  4 23:33:28 client kernel: GEOM_ELI: Encryption: AES-CBC 128
Jan  4 23:33:28 client kernel: GEOM_ELI:     Crypto: software

The device da0.eli has been created - this is the end-point device that can be used by file systems and for other purposes (swap, etc.).

6. Make the file system and mount it!

# newfs -U -L mydata /dev/da0.eli

A successful run of newfs looks something like this:

/dev/da0.eli: 5120.0MB (10485756 sectors) block size 16384, fragment size 2048
        using 28 cylinder groups of 183.77MB, 11761 blks, 23552 inodes.
        with soft updates
super-block backups (for fsck -b #) at:
 160, 376512, 752864, 1129216, 1505568, 1881920, 2258272, 2634624, 3010976, 3387328,
 3763680, 4140032, 4516384, 4892736, 5269088, 5645440, 6021792, 6398144,
 6774496, 7150848, 7527200, 7903552, 8279904, 8656256, 9032608, 9408960, 9785312, 
 10161664

Since we used a volume label for the file system, observe the following message in the kernel log:

Jan  4 23:38:17 client kernel: GEOM_LABEL: Label for provider da0.eli is ufs/mydata.

Now you can mount the file system:

# mount /dev/ufs/mydata /mydata

And that's it!

There are two points that can't be readily automated right now: the iscontrol step which starts the iSCSI initiator, and the geli requiring a password. The former can be approximated by creating a small shell script that does the step and putting it in /usr/local/etc/rc.d but the second cannot be, since the whole point of having an encrypted storage is that it isn't accessible by unwanted people.

The way this setup works is that the unencrypted data is used by the file system (as it should - you wouldn't be able to use it otherwise) via the da0.eli device. This data is encrypted and the encrypted data is written to da0 device. This is the iSCSI client device and the data is tranferred to the server in its encrypted form. The server and the network never see unencrypted data.

Due to GEOM's modularity, other components could be added to the data processing graph, such as journaling (gjournal), caching (gcache), etc. in which case the end-point device name will "grow" suffixes, such as da0.eli.journal. Even RAID levels can be added, though it makes little sense to do it on the client (it's perfectly fine on the server).

FreeBSD gotchas

Apache or PHP crash when PHP is upgraded

This is a known bug in PHP which is exposed by certain ordering of the extensions in extensions.ini. Some of the incompatibilities are documented on php.net, but most aren't. The bug usualy manifets on a newly upgraded PHP when the order of loaded extensions is semi- randomly changed (from a previously working configuration). See this thread for some info.

Apache parent process misteriously dies with PHP with lots of extensions

This one took a lot of time and karma to diagnose. The problem is: apache suddenly misteriously dies, or is in a weird braindead state where every new request cuases the child process assigned to serve it to die.

It seems that the reason for this is in the way FreeBSD handles threading libraries. In particular, FreeBSD has three threading libraries that can be used by any one process (libc_r, libthr, libpthread), in addition to using no threading libraries. As long as a single process uses only one of these options, all is well. The problem is that any single library can (erroneously!) bring a "forced" reference to a threading library, and Bad Things happen to a process that stumbles into such a situation.

My problem was that Apache was built with "prefork" mode (no threading), and one of the libraries used for PHP was bringing in libpthread. This conclusion was arrived to by a trial-and-error process with much educated guessing, since backtraces from the core dump didn't list the library, but it immediately rang a bell - wasn't there a shiny-looking switch in the process of building PostgreSQL's client libraries that said "build libpq threadsafe"? Yes it was. It turns out that libpq was not only being built threadsafe, it was also strongly linked to libpthread.so. This is bad, as it caused Apache to start without threading, then PHP (a DSO module in PHP) would load libpq.so, which would load libpthread.so for that Apache child, which would... somehow... propagate to the parent process and cause it to die trying to call _pthread_main_np function (for which there's only a "weak symbol" in libc and the threading libraries override it with their own implementation). Building libpq.so without "threadsafe" option fixed this.

Apache with PHP or PHP-CGI dies with SIGNAL 11 or SIGNAL 6

This error is commonly found after running recursive portupgrade on the php port. PHP has a long-standing issue with the order in which its extensions / libraries are loaded. On FreeBSD, this order is specified in /usr/local/etc/php/extensions.ini. Running portupgrade will reorder the extensions in an unpredictable way, breaking previously working installations. The solution is to find out which order of extensions works and make a backup of extensions.ini. Searching for "freebsd php extensions problem crash" on Google will find a lot of references to this problem, and the problem is often discussed in FreeBSD mailing lists. Apparently this is not a FreeBSD-specific problem, but it's commonly found here due to the way portupgrade works.

What's cooking for FreeBSD 7?

The next major release of FreeBSD, version 7, is one of the most significant so far, with amount of new technologies and improvement largest since the introduction of 5.0. Since constantly searching the mailing lists for important changes can be a bit tedious, I've created this (frequently updated) page to list some of the more interesting new things in one place.

FreeBSD 7.0 has been released! I've now started the continuation of this project: What's cooking for FreeBSD 8.

Also useful are the quarterly Status Reports:

If you're interested in how FreeBSD gets developed, you're encouraged to read the mailing lists and developer blogs.

Network stack improvements and cleanup

Even though this document mentions only several people, the effort to improve the network stack and its performance has been carried by many.

New `sendfile()` implementation, improved `sosend()`

Status: Committed to -CURRENT
Will appear in 7.0: sure
Author: Andre Oppermann, Robert Watson
Homepage: http://people.freebsd.org/~andre/, announcement message

While working on TSO support, Andre Oppermann has found several ways to optimize kernel's internal networking support. The new sendfile() implementation sends larger chunks of data at once and improves performance up to 5x when used with TSO and other new enhancements. Improvements to the sosend() and related functions resulted in lowering the CPU consumption of sending side of network interfaces almost three times. Note that these are microbenchmarks and performance improvements in real usage still needs to be quantified.

TSO and LRO support

Status: Committed or ready for -CURRENT
Will appear in 7.0: sure
Author: Andre Oppermann and Andrew Gallatin
Homepage: http://people.freebsd.org/~andre/

The ongoing effort to improve FreeBSD's network performance (especially after the hit taken during transition to SMP) has resulted in the new ability to support TSO (TCP/IP segmentation offload) and LRO (Large Receive Offload) hardware on gigabit and faster cards. Some of the drivers that support TSO include: em, bc, cxgb, ixgbe, msk, mxge, nxge, nfe, re (or in plain words: Intel, Broadcom, NVidia, Realtek and other cards, gigabit or better). LRO support is currently in mxge.

TCP socket buffers auto-sizing

Status: Partially committed to -CURRENT
Will appear in 7.0: sure
Author: Andre Oppermann
Homepage: http://people.freebsd.org/~andre/

FreeBSD has a default 32K send socket buffer. This supports a maximal transfer rate of only slightly more than 2Mbit/s on a 100ms RTT trans- continental link. Or at 200ms just above 1Mbit/s. With TCP send buffer auto scaling and the default values below it supports 20Mbit/s at 100ms and 10Mbit/s at 200ms. Both read and write buffer are auto-sized.

While the support for send buffers auto sizing is committed, patches for receiving side are still under testing.

Rapid Spanning Tree Protocol (802.1w)

Status: Committed to -CURRENT
Will appear in 7.0: sure
Author: Andrew Thompson
Homepage: http://people.freebsd.org/~thompsa/

RSTP provides faster spanning tree convergence. The protocol will exchange information with neighboring switches to quickly transition to forwarding without creating loops. The code will default to RSTP mode but will downgrade any port connected to a legacy STP network so is fully backward compatible.

SCTP (Stream Control Transmission Protocol)

Status: Committed to -CURRENT
Will appear in 7.0: sure
Authors: Randall Stewart, George Neville-Neil
Homepage: http://www.sctp.org/

FreeBSD is the reference implementation for the SCTP.

Like TCP, SCTP provides a reliable transport service, ensuring that data is transported across the network without error and in sequence. Like TCP, SCTP is a session-oriented mechanism, meaning that a relationship is created between the endpoints of an SCTP association prior to data being transmitted, and this relationship is maintained until all data transmission has been successfully completed.

Unlike TCP, SCTP provides a number of functions that are critical for telephony signaling transport, and at the same time can potentially benefit other applications needing transport with additional performance and reliability.

Link aggregation / trunking

Status: committed to -CURRENT
Will appear in 7.0: sure
Author: Reyk Floeter (from OpenBSD)
Manpage: lagg(4)

OpenBSD's trunk(4) was imported to FreeBSD in time to be shipped in FreeBSD 7.0. The trunk interface allows aggregation of multiple network interfaces as one virtual trunk interface for the purpose of providing fault-tolerance and high-speed links. The driver currently supports the trunk protocols failover (the default), fec, lacp, loadbalance, roundrobin, and none.

Improvements to kernel facilities

PMC performance monitoring

Status: Available in -CURRENT, partially available in RELENG_6
Will appear in 7.0: sure
Author: Joseph Koshy
Homepage: http://people.freebsd.org/~jkoshy/projects/perf-measurement

This project implements a kernel module (hwpmc(4)), an application programming interface (pmc(3)) and a few simple applications (pmcstat(8) and pmccontrol(8)) for measuring system performance using event monitoring hardware in modern CPUs.

Some parts (hwpmc, libpmc, pmcstat) were developed even before RELENG_6 was branched and new development goals for 7.x include support for callgraphs and a GUI front end.

Interrupt filtering

Status: Mostly committed to -CURRENT
Will appear in 7.0: sure
Author: Paolo Pisati
Homepage: wiki page

Interrupt filtering is a new method to handle interrupts in FreeBSD that retains backward compatibility with the previous models (FAST and ITHREAD), while improving over them in some aspects. With interrupt filtering, the interrupt handler is divided into 2 parts: the filter (that checks if the actual interrupt belongs to a device) and a private per-handler ithread (that is scheduled in case some blocking work has to be done). The main benefits of this work are:

Feedback from filters (the operating system finally knows what's the state of an event and can react consequently).
Lower latency/overhead for shared interrupt line.
Previous experiments with interrupt filtering showed an increase in performance against the plain ithread model in some cases.
General shrink of the machine dependent code - part of the interrupting handling code was turned into machine independent code.

Linuxulator for Linux 2.6

Status: Committed to -CURRENT
Will appear in 7.0: sure
Authors: Alexander Leidinger, Roman Divacky
Homepage: blog post, cvs commit note

FreeBSD includes support for natively executing Linux binaries. This is done via runtime translation of Linux syscalls to BSD syscalls, with no performance penalty. The facility is colloquially called the "linuxulator".

Linuxulator in -CURRENT has been updated to run binaries made for Linux 2.6.16 (though the default for 7.0 will still be 2.4), and the official Linux environment will be Fedora Core 5.

New scheduler: ULE 2.0 / 3.0

Status: Committed to -CURRENT
Will appear in 7.0: sure
Author: Jeff Roberson
Homepage: CVS file reference, commit message, description

The original SCHED_ULE was under-performing and buggy, so it got reworked. The new scheduler replaces, and has the same name as, SCHED_ULE, but is of a somewhat different architecture. It replaces the double queue mechanism with circular queues, and fixes a lot of other things, but it's still an O(1) scheduler with per-CPU queues.

During SCHED_ULE 2 development there was a brief period where there was a third (or fourth, depending on how you count) scheduler, named SCHED_SMP, forked from SCHED_ULE 2 and heavily optimized for configurations with large number of CPUs (8+). This SCHED_SMP has been renamed and committed as SCHED_ULE. While the new scheduler will really shine for multi-CPU machines, it's now also recommended for single processor systems as it has much better interactive performance (mixing of processes with different requirements for IO vs CPU time). ULE will not be enabled by default for 7.0 but it's an officially recommended performance optimization.

Improved accounting file format

Status: Committed to -CURRENT
Will appear in 7.0: sure
Author: Diomidis Spinellis
Manpage: acct(5)

The accounting record format has been revised to store time values with microsecond precision. This allows the recording of meaningful values for short-running commands on modern fast processors. The adoption of the IEEE 754 float format for storing time and usage values greatly increases their range and precision, and also simplifies the processing of accounting records by third party tools. The new record format and the tools lastcomm(1) and sa(8) maintain backwards compatibility with the original accounting format.

Storage subsystems' improvements

ZFS

Status: Committed to -CURRENT
Will appear in 7.0: sure
Author: Pawel Jakub Dawidek
Homepage: announcement message, commit announcement message

Sun's ZFS is in the process of being ported to FreeBSD, with the intention of offering most (or all) features found in the original implementation. It's integrated with FreeBSD's existing features like UFS and GEOM, thus offering the possibility of creating FreeBSD UFS file systems on ZFS volumes, and using GEOM providers to host ZFS file systems.

ZFS is an advanced file system (actually, a combination of file system and volume manager) with many interesting features built-in: snapshots, copy-on-write, dynamic striping and RAID5, up to 128-bit file system size (limited to 64 bits in practice even in Solaris - there's no 128-bit integer type in standard C language), and globally optimal I/O sorting and aggregation. It's marked EXPERIMENTAL in 7.0-RELEASE.

ZFS is still experimental on FreeBSD, and it's recommented that users get familiar with FreeBSD ZFS documentation before using it. For a more light-hearted introduction see this presentation by Pawel.

tmpfs

Status: Committed to -CURRENT
Will appear in 7.0: sure
Authors: Julio M. Merino Vidal, Rohit Jalan, Howard Su, Glen Leeder
Homepage: TMPFS page on FreeBSD wiki, TMPFS at NetBSD

TMPFS is a memory file system designed to efficiently allocate (and deallocate) memory used for the file system itself, as contrasted to the "usual" way of creating memory file systems by creating memory storage devices ("RAM drives"). It's marked EXPERIMENTAL for 7.0-RELEASE.

gjournal

Status: Committed to -CURRENT
Will appear in 7.0: sure
Author: Pawel Jakub Dawidek
Homepage: http://bsdblogs.droso.org/pjd, announcement message

Gjournal is a GEOM storage class that provides data journaling facilities to any providers (and consumers) the user needs. As a special case it has support in UFS file system code, and in this combination it makes UFS a journaled file system. In itself, gjournal consumes two devices (one for the data, one for the journal) and provides one. Since it takes special care to work well with disk drive hardware caches, it can be used to accelerate and provide reliability in many other uses, such as GELI and GBDE encrypted device providers.

I'm proud to say current gjournal is a continuation of my idea implemented for Google's Summer of Code 2005.

gvirstor

Status: Committed to -CURRENT
Will appear in 7.0: sure
Author: Ivan Voras
Homepage: http://wikitest.freebsd.org/gvirstor

Gvirstor is a GEOM storage class that provides a storage device of arbitrary size in "overcommit" mode (i.e. larger than physically available storage). Providers can be added to the virstor device on-line (while used, e.g. mounted), and removed if unused and at the end of the list of components.

This work was created by me, with Pawel Jakub Dawidek as mentor and sponsored by Google in Summer of Code 2006.

gmultipath

Status: Committed to -CURRENT
Will appear in 7.0: sure
Author: Matt Jacob
Homepage: CVS message

Gmultipath allows failover between multiple devices that represent the same storage device. This is an active/passive{/passive...} arrangement that has no intrinsic internal knowledge of whether devices it is given are truly multipath devices. As such, this is a simplistic approach, but still a useful one. The first of N identical devices (and N *may* be 1!) becomes the active path until a BIO request is failed with EIO or ENXIO. When this occurs, the active disk is ripped away and the next in a list is picked to (retry and) continue with.

New platforms

New platform: ARM architecture

Status: Committed to -CURRENT, MFC-ed to RELENG_6
Will appear in 7.0: sure
Authors: Olivier Houchard, Warner Losh & more
Homepage: http://www.freebsd.org/platforms/arm.html, http://bsdimp.blogspot.com/

Support for ARM embedded architecture has been under development since 6.0, enabling FreeBSD presence in the embedded markets.

The support is now MFC-ed to 6.x and is available in 6.2-RELEASE. It's still under development and will likely support more boards in the future.

New platform: sun4v (Niagara / T1)

Status: Committed to -CURRENT
Will appear in 7.0: probably
Authors: Kip Macy, John Birrell & more
Homepage: CVS announcement

There's still a long way to fully supporting Sun's Niagara/sun4v platform, but progress is slowly being made. Niagara offers advanced features such as eight cores and 32 threads per CPU, and hardware public key cryptography acceleration. Unfortunately, this architecture is not supported out-of-the-box in 7.0.

Security features

Security event auditing

Status: Committed to -CURRENT, MFC-ed to RELENG_6
Will appear in 7.0: sure
Authors: Robert Watson & more
Homepage: http://www.freebsd.org/doc/en_US.ISO8859-1/books/handbook/audit.html

Event auditing allows the reliable, fine-grained, and configurable logging of a variety of security-relevant system events, including logins, configuration changes, and file and network access. These log records can be invaluable for live system monitoring, intrusion detection, and postmortem analysis. FreeBSD implements Sun's published BSM API and file format, and is interoperable with both Sun's Solaris and Apple's Mac OS X audit implementations.

Audit framework was MFC-ed to RELENG_6 and is available in 6.2-RELEASE.

New privilege separation capabilities

Status: Committed to -CURRENT
Will appear in 7.0: sure
Author: Robert Watson
Homepage: list announcement

This is a framework which can be used together with MAC to creating policies similar to RBAC (as seen in Solaris & others) which allow the root privilege to be separated into several fine grained capabilities such as "can access the network" or "can bypass file system quotas". This is work in progress and no shipped policy modules directly implement all of the functionality yet.

Multimedia features

Hi-def audio

Status: Mostly committed to -CURRENT
Will appear in 7.0: sure
Author: Ariff Abdullah
Homepage: http://people.freebsd.org/~ariff/HDA/

Newly developed driver, snd_hda has been developed to support professional sound equipment and new hardware on the market. HDA hardware is capable of delivering 192 kHz/32 bit quality for two and 96 kHz/32 bit for up to eight channels. Latency has been reduced for many cases.

Related to this, new drivers for envy24(ht) sound hardware is committed to -CURRENT, and multichannel audio support is due to be finished soon.

Userland enhancements

jemalloc

Status: Committed to -CURRENT
Will appear in 7.0: sure
Author: Jason Evans
Homepage: http://people.freebsd.org/~jasone/jemalloc/

The currently used malloc() library, called phkmalloc since its creator is Poul-Henning Kamp, is almost a decade old in its present implementation. It was designed for a time when memory was scarce, the priorities considered in memory allocation were different, and multithreading was still an academic idea. Even so, it's one of the more popular malloc() implementations, used in all BSDs and even some Linux distributions.

Because of its inefficiency when used in multithreaded applications running on multiprocessor systems, a new userland memory allocator was created, named jemalloc after Jason Evans, its creator. The improvements in computer speed and memory availability mean that compared to phkmalloc, which only needed to be conservative in memory usage, jemalloc needed to be more sophisticated and account for low-level properties such as CPU cache locality and parallel execution.

The result is an allocator which is optimized for multithreading, using multiple allocation arenas to help concurrency. On single processor systems there's only one arena, while on multi-processor or multi-core systems there are four times as many arenas as there are processors. Allocations are divided into broad classes based on their size and those classes are further subdivided. Benchmarks show that jemalloc does significantly better in multithreaded applications (like MySQL) and for applications that make many small allocations.

Bits & pieces

Authors: many

Here are some additional changes for 7.0 that are not so glamorous or are smaller in scope:

Lots of performance improvements on SMP machines (see MySQL read performance, MySQL write performance and BIND performance graphs.)
Significantly increased scalability on SMP machines, mainly from extraordinary work done by David Xu (the libthr threading library), Jeff Roberson (scheduler, flock locking), Atillio Rao (improved kernel locking performance) and Robert Watson (file descriptor locking, unix sockets locking and more).
Significantly increased network scalability, resulting mostly from switch to direct dispatch of the network stack from netisr. This is especially helpful for 10 Gbit/s NICs and was mainly done by Robert Watson and Kip Macy.
GIANT lock has been pushed further back, and almost all kernel subsystems are now finely locked (e.g. VM, VFS, Net). Some of the recent improvements are: locking the CAM subsystem and many SCSI drivers (by Scott Long), and similar locking work has been done on the NFS client and the Firewire implementation.
iSCSI initiator (iSCSI target is available in ports)
SATA support
Read-only access to XFS file systems
Added support for MSI/MSI-X extensions to PCI
Support for Apple (Mac) hardware is being worked on
pf firewall updated to 4.1
X.Org 7.2 - things like beryl now work if you have the right drivers
gcc 4.2
Implemented symbol versioning for many base OS libraries
libthr becomes the default threading library

Things that didn't make it

Despite plans and best efforts, some things won't make it into FreeBSD 7. These are:

SCHED_CORE - Doesn't perform as well as SCHED_ULE2
DTrace - Licensing issues in some Sun's header files are preventing inclusion of DTrace in FreeBSD.
Superpages - Not developed in time to go into 7.0-RELEASE

Of course, this much new technology will need much testing before it's ready for use. You can help by installing a snapshot of -CURRENT and running it on as close to your regular load as possible. Disable debugging features (which are enabled by default during development) before benchmarking.

Predictions of future

It's fun to pretend to predict future events and environments, if only so we can laugh when the predictions turn out to be wrong and much more quaint than the real thing. So, here are my predictions. I don't want to be overly optimistic and put the predictions too near "our time", so let's say they are targeted for about 20 years from now (beginning of 2008.) though some of them may be realized even 10 years from now. Here's the list:

Microsoft will sell cars, or car parts, or fridges. Something "classical".
Dell is the new Apple.
Google is the new Microsoft. Some FutureTech Company is threatening the long and obnoxious rule of Google.
The concept of paying for bandwidth instead of content is relegated to the history books and looks like something from third-world nations in Africa.
Computers without always-online Internet connectivity will be almost useless to the "common man", though there will always be a core of a culture of "old-fashioned" users that want total control over their data (probably evolving from the today's OpenBSD and Crypto nut communities). The "old-fashioned" applications that deal only with local data will stagnate, simply because there's nothing new to invent there (the last real inventions in desktop apps happened in Xerox some decades ago: the mouse and WYSIWYG editors).
A small percentage of "global warming" and other "sky is falling" theories happen. We'll probably lose Venice, but people will continue to live their lives without major interruptions. Possible future "sky is falling" theories include meteor hits and WW3.
USA and Russia are constantly attempting to start another Cold War but nobody cares.

FreeBSD 7 Footshooters

This page will contain a list of (relatively) small annoyances and issues present in FreeBSD 7 (and possibly earlier versions), with the purpose of forewarning and helping others that have stumbled on the same problems as I have. These are nothing special and some of them may be documented elsewhere.

tmpfs and fstab

If you mechanically add "2" to the end of fstab lines (fsck pass#), you'll get a big surprise the next time you reboot the machine if you try to add tmpfs to the list, or change the entry from md-based /tmp to tmpfs. The problem is that there's no fsck for tmpfs so when the machine boots, it will try to check the file system, will fail and stop the boot process. The solution is to simply change the "2" to "0".

SCTP depends on IPv6

You can't build a kernel with SCTP option but without IPV6 option, so remove it together with IPV6.

ZFS manipulation needs writable /boot

Modifying zpools and zfs (e.g. adding devices to zpools) needs to be done with /boot/zfs directory writable. Since /boot is usually on the root file system, the root partition needs to be writable. This is important in single user mode where the root file system is mounted read-only. After mounting it rw, and before starting ZFS with /etc/rc.d/zfs start, hostid must be initialized with /etc/rc.d/hostid start. Failure to have /boot writable while changing zfs structure will result in a panic when ZFS is started "regularily" the next time. This can be fixed by deleting zpool.cache from /boot/cache and performing zfs import. Failure to initilize hostid will result in ZFS file system(s) not being present the next time ZFS is started, which can be fixed by zfs import -f.

Compiler flag -Os might produce bad code

There have been many reports, and my own experience supports them, that using gcc -Os flag (optimize for smaller code) sometimes produces bad code. The symptoms vary wildly over architectures and flag combinations, so that a system that appears to be working fine might have subtle bugs lurking to be discovered in the most inconvenient time. This affects both the kernel and the userland. The compiler in FreeBSD 7.0 is gcc 4.2.1. This might be fixed in the future by importing a newer version of the compiler, but don't assume it's happened if you need the flag.

ZFS is only experimental

ZFS (the file system) is an experimental feature in FreeBSD. Don't even try using it before proper tuning.

Conferences

I was present and presented the following papers (or gave named lectures) at various conferences:

BSDCan 2007 Developers' Summit talk about finstall: ivoras_finstall.pdf (BSDCan)
BSDCan 2007 Developers' Summit talk about gvirstor: ivoras_gvirstor.pdf (BSDCan)
Network Distributed File System in User Space tdfs.pdf (ITI 2006.).
A hierarchical file system interface to database-based content management application ferwebdav.pdf (ITI 2006.).
On recording and presentation of measurement data acquired via web services webservicesdata.pdf (MIPRO 2006)
Integracija Jabber sustava u CMS DORS2006_Jabber.pdf (DORS/CLUC 2006.)
A WebDAV view of FER's Content Management System ivoras_ferweb-webdav2.pdf.gz (CUC 2005)
Distribuding a Web-based Content Management system distribferweb_iti.pdf (ITI 2005)
WebDAV u uredskom okru�enju webdav.pdf (DORS 12, 2005)
FreeBSD 5.2 freebsd52.pdf (DORS 11, 2004)
Kako ubrzati PHP program ubrzatiphp.pdf (DORS X, 2003)
FreeBSD 5 - �iji je Unix dulji freebsdlinux.pdf (CLUC V, 2003)

Except if otherwise specified, these documents are for personal use only. No printing, lecturing or distribuding them is allowed, or any other use for commercial purposes is allowed without written consent from the Author.

Osim ako nije druga�ije nazna�eno, dokumenti se smiju koristiti samo za osobne potrebe. Nije dozvoljeno umna�anje, prezentiranje ili kori�tenje u komercijalne svrhe bez pismene dozvole autora.

S e a r c h

Tree macro tutorial

BSD systems include macros for several useful structures and algorithms, including several types of lists and trees. While the lists are easy to use, I always forget the right order of declarations for trees. So here it is, how to use tree macros in FreeBSD (C language):

First, declare the structure that is to be stored in the tree:

    struct mydata {
        RB_ENTRY(mydata) linkage;
        int payload;
    };

Second, declare the comparison function. This function compares two structures in the way similar to strcmp():

    static int mydata_cmp(struct mydata *e1, struct mydata *e2) {
        return e2->payload - e1->payload;
    }

Next, declare the head structure and head entry. Head structure is the struct type of the tree head, and it's the way tree is accessed.

    RB_HEAD(mydata_entries, mydata) head = RB_INITIALIZER(&head);

You're now ready to declare the prototypes for the internal tree structures and the functions themselves:

    RB_PROTOTYPE(mydata_entries, mydata, linkage, mydata_cmp);
    RB_GENERATE(mydata_entries, mydata, linkage, mydata_cmp);

The tree can now be used normally, the way described in the manual:

    struct mydata *data;
    RB_INSERT(mydata_entries, &head, data);

    struct mydata find;
    find.payload = 42;
    data = RB_FIND(mydata_entries, &head, &find);

    RB_FOREACH(data, mydata_entries, &head)
        printf("%d\n", data->payload);

In the examples above:

mydata is the structure to be stored in the tree. It contains some arbitrary payload data but must contain an TREE_ENTRY element.
mydata_entries is the type that contains the tree. It's declared by the RB_HEAD macro.
head is the tree head.

Papers

Academic/scientific papers

Title	Abstract	Event
A distributed system for data storage and retrieval (in Croatian)	This work studies technologies applied in building distributed peer-to-peer network systems for data storage and retrieval. These systems consist of arbitrary number of network nodes connected on a higher layer and their connectivity is not influenced by physical topologies of networks of which they are a part of - they form "overlay networks." All nodes in a peer-to-peer overlay network have equal functionality (there are no "server nodes"). This equality of nodes presents some unique problems: mutual discovery of active nodes and varied connectivity as nodes join and depart the network. Various solutions for these problems are presented and evaluated and an architecture for a peer-to-peer system is proposed. The proposed system uses TCP/IP protocols for inter-node communication and allows storage and retrieval of data records in the form of dictionaries (mappings). A prototype of the system is implemented in Java programming language, and technical and user documentation is provided in this work.	dipl.ing. thesis
Network Distributed File System in User Space	File systems have traditionally been implemented in the operating system's kernel to ensure maximum possible speed and integration with the rest of the operating system, and this was true even for network file systems such as NFS. However, available CPU power on mainstream architectures continues to increase daily at a rate which is not closely followed by speed of computer network equipment. When considering development of network-distributed file systems today it becomes clear that speed improvements offered by pure kernel-side implementations are no longer significant given the bandwidth and latencies of computer networks. Recent efforts in enabling user-space file system implementations on free / open source Unix-like operating systems have made it possible to create a solution for distributing file system data over computer networks entirely in user-space. In this work we present such a solution - the Trivially Distributed File System.	ITI 2006
A hierarchical file system interface to database-based content management application	When considering the usability of computer applications one, of the most important factors is the interface they provide for data manipulation. Historically, though each application has defined it's own user interface, some types of interfaces have emerged as most applicable for certain types of data. Over time, some applications (or types of applications) have gained enough popularity that they became ubiquitous and well-known to most computer users to the extent that users consider them the norm, and as such, optimal for their daily tasks. In this paper we present an idea and implemented method of exposing data from a web content management system in the form of hierarchical file system, manageable and editable by usual file management and office application tools.	ITI 2006
On recording and presentation of measurement data acquired via web services	Web services have greatly eased exchange of data, but the usage of web services carries its own specific downsides as well as advantages. Most notable of these are latency of transmission and reliability of server equipment. In this work a straightforward solution for distribution and representation of measurement data was created. This solution is comprised of two parts: a back end process which communicates with the data source via a web service protocol based on XML (in Python language), and a front-end module for the FERweb Content management system (in PHP language). These processes communicate asynchronously via the database.	MIPRO 2006
Distributing Web-based Content Management System - "FERweb"	This work explores various ways of distributing the FERweb CMS system (a web-based Content Management System of the University of Zagreb, Faculty of Electrical engineering and computing) Primary aim of this project is enhancing the performance of the system, within the constraint that the efforts must be based on existing technologies used in the project and with minimal impact on the existing code.	ITI 2005

General papers

Title	Date	Abstract
On implementation of `switch()..case` in gcc	January 2 2005	A simple exploration of how the `switch()..case` construct is compiled into machine code in gcc 3.4
Distribuding Web-based CMS	June 24, 2005	This is the "official" version of the following paper, presented at the ITI 2005 conference.
Web CMS Distribution	May 7, 2004	This work explores various ways of making the FERweb CMS system (a web-based Content Management System of the Faculty of electrical engineering and computing at the University of Zagreb) a "distributed system". Here, the notion of a distributed system is taken quite literally, and stands for executing as many components of the system as possible on as many different computer systems. The only limits here imposed are that of the practical value of such distribution . its end result must be either an increase of total performance delivered to the user of the system, an increase of overall system reliability, or providing distinct new functionality that was previously not possible, and that the efforts must be based on existing technologies used in the project already, with minimal impact on the existing code. Where appropriate and where available resources permit, measurements are taken and reported.
OS Benchmark	April 10, 2004	This article presents the results of benchmarking various operating systems. The benchmarks include synthetic tests (bytebench, ubench, bonnie++) and, more importantly, semi-"real world" tests (pgbench, web CMS). Operating systems benchmarked are various versions of FreeBSD, DragonflyBSD, NetBSD and Linux.
GCC as Optimising Compiler	Feb 17, 2004	This paper aims to present and analyse some of the optimisations that are conducted by the GNU C compiler, in version 3.3.4 (on a i386 FreeBSD system). This is only an informal and introductory paper and is not meant to be comprehensive nor complete, only to give insight at things that are happening ``behind the scenes''.
Description of PHP-RPC protocol	Jan 15, 2004	Because of the need for a truly light-weight RPC mechanism in PHP, a new protocol is formed: PHP-RPC, to be used instead of XML-RPC when communicating to purely PHP applications. This protocol is faster and has a smaller overhead.

Curriculum papers (mostly in Croatian)

Title	Date	Abstract
RDRAM Memorija	Feb 8, 2004	Ovaj rad opisuje osnovni rad Direct Rambus RAM sustava, kakav se jo� uvijek koristi. Na kraju rada je dan kratak opis promjena u novijim ina�icama sustava.
KSE sustav u FreeBSD 5.0	Dec 2002	Cilj ovog rada je prikazati mogu�nosti postizanja vi�edretvenog rada u operacijskom sustavu FreeBSD 5.0, sa posebnim osvrtom na novorazvijeni sustav KSE: Kernel Scheduled Entities koji donosi mnoga pobolj�anja u odnosu na prethodno dostupna rje�enja, posebno u podr�ci za iskori�tavanje vi�e sistemskih procesora (ukoliko su prisutni) i konkurentnog izvr�avanja poziva jezgre operacijskog sustava. Prakti�ni dio rada demonstrira kori�tenje razli�itih postoje�ih metoda postizanja vi�edretvenosti u aplikacijama, te njihove posljedice za aplikaciju i cijeli sustav.

Other (conference presentations, etc.)

Title	Date	Abstract
FreeBSD's finstall	May, 2007	Presentation of new FreeBSD graphical installer, given at BSDCan 2007.
FreeBSD's GVIRSTOR	May, 2007	Presentation of GEOM_VIRSTOR kernel module for FreeBSD, givent at BSDCan 2007.
WebDAV u uredskom okru�enju	April, 2005	Kori�tenje WebDAV-a u poslovnom okru�enju, primjene, poslu�itelji, klijenti.
FreeBSD 5.2	April, 2004	FreeBSD arhitektura i okru�enje, novosti u 5.2, napredne mogu�nosti uporabe
Kako ubrzati PHP program	Mar, 2003	PHP je spor! Zbog prili�no "labave" sintakse, PHP interpreter mora prolaziti kroz puno vi�e provjera nego za neke druge skriptne jezike. PHP-ov "model programiranja" je �esto neoptimalan s resursima...
Uvod u FreeBSD	Mar, 2003	�iji Unix je dulji? Kratka povijest vremena, BSDL vs GPL, BSD is dead! *BSD, Prednosti FreeBSDa...

There's more!

There's a separate section containing my FreeBSD papers.

Should I buy a Mac?

A Mac?

Thanks to a friend, I had a chance to try two Apple laptops, a MacBook and a MacBook Pro over the weekend. The reason for this is that I wanted to try a Macintosh system for some time now, but I never had a chance until now. I mean, I saw the commercials, the web sites and the blogs, had a chance to tinker with the occasional demo system but I never used them for more than 5 minutes.

I'd first like to vent some acid about the availability and pricing of Apple computers in this sorry little excuse for a country. It seems there's only one Apple partner/reseller in this country, and the prices they have are very unreasonable. The MacBook costs (when directly converted back to USD) around $2100, and the MacBook pro costs close to twice than that, around $4100. Compare this to regular Apple prices (in US) and you'll see the local prices are doubled! I know Apple computers are traditionally priced up to 25% more than equivalent PC laptops but this is ridiculous.

Hardware

By the feeling of it, both laptops seem to weigh about the same (I didn't measure it with scales), which is suprising for the smaller one. The MacBook is too heavy for its size, especially when compared to similar PC laptops. I don't know who designed the keyboard on the MacBook nor why he designed it like that, but he should be shot for it. Not only is it ugly, but it's also unergonomic in several ways: the keys have an awkward shape and angle (totally parallel with the surface of the laptop), and the space between the keys is too large for comfort. The keyboard on the MacBook Pro looks small for the size of the laptop surface, but the keys are actually large enough. There's plenty of empty space around it could maybe be filled with more buttons or controls (the "function" keys are half-height and double for hardware control keys like audio volume and screen brigthness - they could have been larger and the hardware controls could have been separate). The quality and feeling of it is much better than that of MacBook but it's still behind that of quality PC laptops. The lid of the MacBook Pro is opened by a small narrow "button" which is tedious to push as it's too narrow for my fingers. Overally, both models "feel" more robust than average PC laptop hardware, with the only exception being the CD loader, which emits strangely screeching noises when loading and unloading CDs. The location of the CD/DVD slot drive on MacBook Pro - at the front of the laptop - is a bad joke. The only hardware-related thing I'm impressed with is the magnetic power plug - it's much cooler that I thought it would be.

The screens on both are reasonably good but again not as good as on high end PC laptops (especially Sony's). On the MacBook the colors seems somewhat washed up (noticable while playing DVDs).

The connectors on the MacBook are badly placed - all of them are on the left side of the laptop, placed too close together, which is again unreasonable as there's plenty of empty space on that side alone. The MacBook has only two USB ports, one next to the other, which creates problems if you have bulky USB devices or connectors. MacBook Pro, though much larger, has almost the same number of ports. There are still only two USB ports, but luckily this time placed on both sides. The network port (RJ-45) is on the right hand side now, which is better since most people are right handed. There's no VGA port (there's an adapter), which makes both models unreasonably tedious to use for giving presentations over a projector.

Software

Keyboard layout deserves a special mention. As a person that's been using PCs and PC laptops "forever", I think some of the choices Apple made are simply not reasonable. It's a special problem for me as I'm using non-US layout (Croatian), and while the layout is close enough to that used on PC laptops, there are three big problems with it: the first is that the "Y" and "Z" keys are switched (this is a special local problem here as they have been switched before by the industry in early nineties and people have just gotten used to it - it's a small matter but it creates stupid typos). The second is that the "special" keys or puncuations that have been replaced by local diacrytic characters on the keyboard can be accessed by completely different key sequences than on PC keyboars. For example, to type the "@" character on a PC keyboard, the sequence is AltGR-V, but on a Mac it's Ctrl-Shift-2 (three keys!), which is almost impossible to type with one hand. The "|" character on PC is AltGr-W, and on the Mac it's Alt-Shift-�, which is not consistent with the sequence for "@". The third problem is that the "Enter key" is much too small (most of it is about 1 cm in width). All these problems are related to the local version of the Mac keyboard and probably don't exist for US users on the US keyboard, but are very annoying for people who work with both PCs and Macs over here. As I do a lot of writing (both programming and "normal" text), this is simply a handicap.

After a short time to get used to it, MacOS X (10.4) seems much better than the hardware it runs on. Though it doesn't look as impressive as it does on commercials and demos, it's decent and pleasent to work with. I have the feeling that its graphic capabilities are underused. Maybe it's because of new developments in Linux and Windows, but it looks somewhat simplistic in comparison. I can find four major things I dislike in the UI: the buttons don't have enough visual feedback when they are pressed, it looks like there's a bug that "locks" windows in drag mode even when drag-lock is disable, the mouse pointer controlled by the track pad doesn't continue to move in the same direction when you reach the end of the trackpad with your finger like it does in Windows and Linux, and applications are incisistent in how they look ("brushed metal" vs "light-gray gradient" skin). Overall I'd say these are really minor problems and I think I can get used to them. The Spotlight feature is very neat, but I didn't create nearly enough documents to test it fully.

The overall philosopy of the UI is fine, I'd say the applications I've tried behave more consistently than on Windows or on Linux, so this is a definite good point. Unfortunately, I didn't try XCode as I hoped I could because the keyboard layout is just depressingly hard to use for programming.

I managed to find all applications I'd need in normal use, either as bundled or on the web (the fact that OS X is Unix underneath has helped this a lot). At first I thought I would have objections to the "single-mouse-button" policy, but I got used to it very quickly. I'd still recommend getting a two-button mouse for real work but it's not that bad even with the single button.

Hmmm...

At the end, I decided not to buy any of them. The MacBook Pro is so expensive that it's silly (at local prices), and the MacBook has too many hardware issues (bad placement of ports, keyboard that seems to be designed to make your hands hurt while using it, unusually heavy weight) to be usable. I think I can see myself using a MacBook Pro if, for example, someone bought it for me or I win it at a lottery, and I find a way to fix the keayboard layout issues, but I just don't feel like buying it for my own money. The "wow" effect is practically nonexistant.

Broodings and daydreams

This is just a collection of random thoughts. Don't take them seriously.

BSDCan 2007 and FreeBSD DevSummit 2007

BSDCan

It's been a very interesting and useful conference, I've learned much from it and met many interesting people. It's no use just listing the talks given on the conference (you can look that up here), but, together with DevSummit, this was one of the best events I've ever been to.

I'll try and think of something more to say...

DevSummit - Day 2

This is the "official" day of DevSummit talks, and has so far proven to be very interesting and instructive. Reports on ZFS and SMP scalability were also uplifting: it's really great to see so much good work being done recently. An interesting piece of information was the result of recent scalability measurements: FreeBSD 4.x run excellent on 1 CPU machines, 5.x scaled to 2 CPUs, 6.x to 4 CPUs, and 7.x will run great on 8 CPU machines (see here).

Another very interesting talk was a self-referential, meta-presentation on giving presentations about FreeBSD by Robert Watson, which also addressed the problem of having too much to say and ttoo little time in a... unique... way :) During the talk he brought up a very interesting point about the way FreeBSD is organized: it's one of the very rare open-source projects that have institutionalized "transition of power" - leadership is reelected every few years and is in no way fixed or messianic.

This was also the day I gave two of my talks, about gvirstor and finstall!

DevSummit - Day 1

Informal sessions were spread across the day, and have proven to be very interesting. I was mostly on the "alternative" track, participating in discussions about the ports tree and storage systems. Judging from the reaction a brief mention of it provoked, it looks like finstall is going to be very popular!

DevSummit - Day 0

The flight was long and the customs check barbaric, but finally I've arrived at University of Ottawa campus to attend the DevSummit and BSDCan events held there. The campus is really nice and big, I really like the architecture.

Since it was late afternoon, the only thing I could do was join the others at the highly regarded introductory beer consuming at local Royal Oak pub.

Roadrunner and the unfairness of life

I doubt I'm the first one to notice it, but (re)watching the old "Road runner" cartoons it struck me just how unfairly the life treats Coyote (as well as how brutal and cheerfully violent the cartoons are - no way something like that would be made nowadays). Like all kids, I enjoyed watching Wile E. Coyote fail in comical ways, getting kicked, squashed, burned and blown up, laughing all the way.

The whole cartoon is the embodiment of the "Life isn't fair" line. It just isn't fair (to coyote)! Starting from the cartoons' name: everyone knows them as the "Road runner cartoons" despite that 90% of the time we look at the Coyote character, starving and trying to survive by catching the scrawny (and stupid) bird. And it's not like the Coyote is being melodramatic or depressed - life actually isn't fair to him, and everything has turned against him. The laws of physics have been rigged so he will suffer: he's attracting heavy, sharp, hot and explosive objects like a multi-purpose magnet (or a temporary black hole) , the Roadrunner can walk on thin air but that same air will drop him to his doom, explosives will not explode except when he's near them, slings and giant springs have Newton's second and third laws suspended until he's on the path of maximum damage (to himself), and energy sources (batteries and fuel tanks) will only (but always) fail when his vehicle is upside down. The Roadrunner apparently has telekinetic and ESP abilities allowing it to sense his weapons and redirect their fire to the Coyote, and they will hit him no matter where he hides.

The Coyote is like some modern Tantalus, tortured forever by the near presence of his food but never allowed to eat it. Alternatively, maybe he was Hitler in his former life and his karma is just catching up with him. There's never any blood on him or around him - the worst that happens is that his body gets mangled and his fur burned, so he continues to live forever (probably in pain).

On the other hand he sometimes really is stupid, making his experiments under unstable boulders, trusting his life with untested contraptions and, at last, failing to find something else to eat. He can also be considered the embodiment of stupid optimism.

Python Blowfish

This is a slightly modified version of the pure python implementation of the Blowfish cipher by Michael Gilfix. This version contains methods to encrypt and decrypt arbitrary strings (of arbitrary length) using the CTR cipher mode. I've added three new methods: initCTR(), encryptCTR() and decryptCTR() to make this happen.

Download Blowfish in Python module.

Ideas

I'm an ideas kind of guy - I tend to have a lot of ideas and no time to implement them. Since ideas are a dime a dozen, and there's no way I can find the time and resources (i.e. money) to make use of most of them, I'll just dump them on this page. Of course, most ideas are garbage, and the same holds for those on this page.

The small print: If anyone actually manages to implement an idea presented here, I'd appreciate I get mentioned in credits (also, I won't refuse money :) ). If anyone wants to implement an idea on this page and really wants me to remove the idea from the page, I can be persuaded (let's trade).

Anyway, here's my idea dump, in no particular order:

Hardware

Oval displays? In the almost-traditional over-the-top "what will Apple do next, designer-vise" discussion with a friend we came up with an "interesting" idea - round / oval displays. Of course, it's a stupid idea. Or is it? Technologically, an OS that has such advanced display features as OS X shouldn't have any problems with maintaining an oval display region / desktop. The one major technological problem could be waste area during production, if the displays are to be "cut out" of a rectangular slab of LCD. That leaves us the UI / human interface problem. Does it have any benefits here (aside from being unusual and sexy given current technology design standards)? The human field of vision is oval, so it could be argued that an oval display is more adjusted to it. Except for blocks of text, I believe that most other design elements can be modified to fit an oval shape.

2x 2.5" drives in a 5.25" bay I want a 5.25" case (internal HDD enclosure) like it's used for holding 3.5" hard drives in a 5.25" bay, but one that holds two 2.5" drives (one above the other), with proper cooling and hot swap ability. I want to do create a RAID1 array from the drives (hardware RAID optional, but forget it if it's not smart enough to handle hot swapping properly). Computers usually have more than enough 5.25" bays so why not use them?

"Auxels" The problem: Star-trek interface (i.e. touch screen) is cool, but the ape descendants using them (e.g. people) like tactile feedback. I think this kind of feedback could be provided by low frequency acoustic vibrations from behind the screen, Since these must be localized to where the user "clicked" with his finger, the audio sources or conducts must be spread across the surface, pixel-like. Hence "auxels" - acoustic pixels.

Software

JavaScript application server Current web development environments as a rule, use a different language at the server side (e.g. PHP) and at the client side (JavaScript). It would be great to use a single language (by default - JavaScript) for both, thus dropping the requirement that programmers need to use two languages and allowing much better client-server integration (i.e. AJAX-y features). This is something I actually intend to implement, if I ever get the free time.

JavaScript application "platform" Creating basic user interfaces in HTML is fairly easy when compared to any other method except visual "drag and drop" design. At the same time, HTTP's model of transactions (do UI on the client side, then POST the data to the server for processing) is easy and simple to understand. Why not create a sort of embeddable server which would allow developers to code for the same processing model (i.e. HTML or XML templates, business logic code, backend code, database) in plain JavaScript and package it all in one desktop application ("exe")? I don't consider desktop applications dead yet.

P2P mp3 streaming Many people within the office listen to MP3 streaming "radio stations", and many of them listen to the same ones. Wouldn't it be great if the clients themselves became sort of repeating stations? Each client listening to a stream could offer the same stream to the other computers in the office, possibly via mDNS or similar discovery protocol. In effect, this would save enormous bandwidth on the Internet.

This could as well be generalized to other types of traffic, such as web pages. BitTorrent is a right step in this direction.

Imaginary RPG Imagine a CRPG which is not much tied to the physical world. In it, events and environment depend on the imagination and the "willpower" of the participants. Let something like Solaris (from Stanislav Lem's book) or Limbo (from Planescape RPG setting) grant participant whatever they wish for, and let it be real. Let people have their wishes granted, and get in conflict with the wishes of other people!

I guess I'm after a sort of game with tangible (to make a pun) duality between the real and the unreal. If a man walks the desert and wishes it, he could perceive the desert as a forest, and as far as he is concerned, it WOULD be a forest. But unless he's powerful enough, a wall in real world will appear as wall in his own reality.

Or maybe a more interesting thing would be "reality by consensus", where actions and the environment up to a point in future are implicitly or explicitly "agreed upon" by the participants, and everyone's wishes have a certain echo in the reality-to-be.

This looks like a primarily non-combat CRPG, but there's space for combat too - I'm thinking of something like the ending combat in the Star Trek novel Spock must die, where the combatants "wish" various (mostly environmental) cataclysms on each other.

SciFi/stories

Occasionally, I just get random ideas that are best classified as SciFi stories' plots :)

Loophole for FTL? Photons have momentum but zero rest mass. What if you try to STOP photons as-is, would you get infinite energy?

A non-linear universe Consider how light spreads through glass: photons are constantly absorbed and released by the atoms in their path. Suppose that releases are not strictly in the same direction, but *always* vary, either constantly (always x deg from the entry point in a certain direction) or randomly. Could lens-optics be possible? Is it happening now? By a very slight amount related to the alpha constant?

Blinded by the sight of God? God said in the Bible: "no man shall see me and live!" - what if God is everywhere and everything, and this is the reason for mortality? Would the sensory deprived from birth live forever?

Perpetual catastrophe Oldish couple looking at sunset over a skyscraper city. So beautiful, so fragile. It's discovered that the galactic core has exploded (with a nod to Larry Niven), and the shock wave is coming just behind the light (or some other random catastrophe - volcanic winter?). Most people are despairing and rowdy, some (the couple is one of them?) are building a room - cubic room with many layers of shielding, to contain a monument (a stone slab) with writings, and cultural treasure around it. The room's shielding is a marvel of technology. Out of boredom, someone sets a satellite/s searching for the room signature and finds there two almost identical sites! When dug out, the second room looks almost destroyed, but in there are writings in it that look like they talk about sudden discovery of "the second one", which are dismally copied again, on the newly finished monument, before it is sealed.

Nagging suspicion "There's something sinister when a accomplished and fruitful writer receives, every week precisely, one of his own books by mail." Each book is neatly wrapped in decorative paper, with a paper card containing the preceding sentence attached with a piece of plain string to the wrapping.

(The writer goes mad and kills the postman :) )

O Fermi where art thou? What is the easiest way for a long-living (immortal even) race to discover, in a galaxy full of stars each of which could have it's own system of planets, where the best mining areas, rich with needed (and expensive) substances are? How about this: bio-engineer an evolution, starting with several thousands of possible life-starting structures, and culminating one day in beings marveling at their intelligence when they discover radio, and seed the galaxy with it. After that, they'd only have to listen for beacons appearing, beckoning from spots across the space.

The late race The cosmic background radiation is low, around 3 kelvins. Imagine how much time does it take for a universe created in a Big Bang to cool down so much! Regarding the Fermi paradox: this universe may be dying - all the civilizations have come and gone, and those who remain would be saddened and depressed looking at the night sky that is mostly black. Our civilization is mostly dark, almost depressed and disinterested - were the youngest civilization happier, sparkling with the enthusiasm of a newborn universe?

Does the Matrix have debuggers? Hutchison Effect (various weird Tesla+Philadelphia experiment+anti-gravity effects) was not repeated by Hutchison since 1980ies - he claims that it simply doesn't work now. What if something has changed since the time he did it (and possibly when Tesla did it) and now? Was a bug fixed in the Matrix?

Zero point energy? Zero point energy of the vacuum has been proven and theoretically calculated. In relativity, energy and mass are almost interchangeable. So zero point energy, which has extremely huge densities, could "produce" enough gravity to visibly bend space. So far, no fiction. But we're not noticing any space-bending in random blocks of vacuum. What if matter and its effects we notice is actually the LACK of (zero-point) energy?

A realistic Matrix Imagine a world heavily interwoven with digital information and presentation-enhancing technology, for example where words in books get highlighted when people touch them, automatically bringing on a view of encyclopedia that defines it, etc. People are happy and always young. Then a reality is revealed: it's a Matrix-like world where people painstakingly save money IRL to have a few hours of virtual happiness, and IRL live in dirty conditions in what are practically sleeping pens not homes. (Hmmm... cf. Gibson's Matrix)

Randomness as a positive evolutionary trait? Some "Dark Enemy" or Force or whatever is preying on all intelligence, it's attracted to intelligent creatures and kills them/turns them into itself/does something horrible. It is fought, defeated and exiled into another dimension/galaxy/some dark place thought uninhabitable. Certainly, no intelligence would arise in such a place. But it does, and eventually is attacked by and defeats the Dark Enemy, by evolving to be as much random as possible, never going in straight lines, always doing multiple actions at the same time, which seem unrelated and purposeless, so it isn't perceived as intelligent behavior. Its actions result in a Rube Goldberg-style "incredible machine" sequences with far-reaching consequences. Then it goes exploring "Our Place"... would we notice before it's too late?

Security by obscurity In ancient books, names of demons and other Scary Things were often deliberately misspelled in order not to provoke their manifestation by accidentally speaking such a name. The practice went on for a long time and at the end no one really knew the right names to speak. What if some loser finds such a book, tries to pronounce a name for fun and accidentally stumbles on the right pronunciation? Instant dark ages - a demon appears, all science suddenly stops working or malfunctions...

ivoras' Home

FreeBSD under VMWare

1. General tips

2. Don't use lnc

3. Reduce kern.hz

4. Disable internal VMWare swapping

5. VMWare Tools not necessary

What's cooking for FreeBSD 8?

Kernel & low level improvements

MPSAFETTY

Kernel memory limit on AMD64 increased to 6 GB

Kernel threads

procstat(1): A process inspection utility

TextDumps: gathering information after kernel panic

ULE 3.0: New version of the SMP-optimized scheduler

Superpages

DTrace

Networking improvements

VirtNet / VIMAGE / Imunes / Network stack virtualization

Multiple routing tables / FIBs

Equal cost multipath routing

Zero-copy BPF

Kernel NFS locking support

Storage subsystems' improvements

Boot support for GPT partitions

bsdlabel gets extended to 26 partitions

Other changes

My CV

My FreeBSD things

What's cooking for FreeBSD 8?

Miscellaneous

Encrypted iSCSI device tutorial

Setting up an iSCSI target

Setting up the iSCSI initiator

FreeBSD gotchas

Apache or PHP crash when PHP is upgraded

Apache parent process misteriously dies with PHP with lots of extensions

Apache with PHP or PHP-CGI dies with SIGNAL 11 or SIGNAL 6

What's cooking for FreeBSD 7?

Network stack improvements and cleanup

New sendfile() implementation, improved sosend()

TSO and LRO support

TCP socket buffers auto-sizing

Rapid Spanning Tree Protocol (802.1w)

SCTP (Stream Control Transmission Protocol)

Link aggregation / trunking

Improvements to kernel facilities

PMC performance monitoring

Interrupt filtering

Linuxulator for Linux 2.6

New scheduler: ULE 2.0 / 3.0

Improved accounting file format

Storage subsystems' improvements

ZFS

tmpfs

gjournal

gvirstor

gmultipath

New platforms

New platform: ARM architecture

New platform: sun4v (Niagara / T1)

Security features

Security event auditing

New privilege separation capabilities

Multimedia features

Hi-def audio

Userland enhancements

jemalloc

Bits & pieces

Things that didn't make it

Predictions of future

FreeBSD 7 Footshooters

tmpfs and fstab

SCTP depends on IPv6

ZFS manipulation needs writable /boot

Compiler flag -Os might produce bad code

ZFS is only experimental

Conferences

S e a r c h

Tree macro tutorial

2. Don't use `lnc`

3. Reduce `kern.hz`

New `sendfile()` implementation, improved `sosend()`