The arrow of time

Here's what unixbench, a mostly single-process benchmark (all except shell scripts) can get on raw dual Xeon 5430 (2.6 GHz, 8 cores total, amd64 in all cases):

TEST                                        BASELINE     RESULT      INDEX

Dhrystone 2 using register variables        116700.0 15517961.7     1329.7
Double-Precision Whetstone                      55.0     3503.0      636.9
Execl Throughput                                43.0     2093.3      486.8
Pipe Throughput                              12440.0  1258069.1     1011.3
Pipe-based Context Switching                  4000.0    52388.3      131.0
Process Creation                               126.0     3569.8      283.3
Shell Scripts (8 concurrent)                     6.0      733.0     1221.7
System Call Overhead                         15000.0  1011896.9      674.6

And here is the virtualized result:

                     INDEX VALUES            
TEST                                        BASELINE     RESULT      INDEX

Dhrystone 2 using register variables        116700.0 15002297.7     1285.5
Double-Precision Whetstone                      55.0     3511.2      638.4
Execl Throughput                                43.0      512.6      119.2
Pipe Throughput                              12440.0  1077539.8      866.2
Pipe-based Context Switching                  4000.0    30414.4       76.0
Process Creation                               126.0      641.1       50.9
Shell Scripts (8 concurrent)                     6.0      111.3      185.5
System Call Overhead                         15000.0   680753.7      453.8

As expected, anything involving a lot of syscalls or other context switches will suffer. This is mostly disk and network IO, but it's surprising how low the concurrent shell scripts test gets when the number of vCPUs increases.

On the other hand, it really *is* more convenient this way...