HTTP performance compared to VMware ESXi

[Joel Carnat <joel%carnat.net@localhost>]

Hello,

I'm on my way to replace my personal LAN from VMware ESXi to Xen (now  
that Xen supports speedstep :-)
I do this for two main reasons ; first is that I prefer to run NetBSD  
(hence NetBSD/xen) ; then I felt like my NetBSD virtual machines were  
a bit slow running in VMware. I just couldn't mesure it. It was just  
an impression. Windows Server VMs were faster on VMware than on Xen  
though.

Anyway, I'm ready to switch so I have both environment running and ran  
a few 'ab' command (ApacheBench/2.3) from my Mac workstation.
And the results are really different.

I ran "ab -n 500 -c 10 -k" on a dokuwiki webpage (PHP5) that parses a  
few RSS feeds (from Yahoo Pipes) and present them on the Web page.

The VMware instance (netbsd-4, 512MB of RAM), running lighttpd/1.4.20,  
gave the following results:
Concurrency Level:      10
Time taken for tests:   57.070 seconds
Complete requests:      500
Keep-Alive requests:    476
Total transferred:      11851353 bytes
HTML transferred:       11606474 bytes
Requests per second:    8.76 [#/sec] (mean)
Time per request:       1141.394 [ms] (mean)
Time per request:       114.139 [ms] (mean, across all concurrent  
requests)
Transfer rate:          202.80 [Kbytes/sec] received

The Xen domU (netbsd-5, 512MB of RAM), running lighttpd/1.4.22, gave  
the following results:
Concurrency Level:      10
Time taken for tests:   18.929 seconds
Complete requests:      500
Keep-Alive requests:    1
Total transferred:      11857528 bytes
HTML transferred:       11626500 bytes
Requests per second:    26.41 [#/sec] (mean)
Time per request:       378.575 [ms] (mean)
Time per request:       37.857 [ms] (mean, across all concurrent  
requests)
Transfer rate:          611.75 [Kbytes/sec] received

I also tried with a static content page (although generated via PHP).  
The difference is slightly less but still there (34 RQ/s vs 19 RQ/s).

I have tuned the VMware instance to use a wm0 interface, but still  
then, the test show a really huge difference.
Would this be due to the difference between netbsd-4 and netbsd-5 or  
did one notice the difference too between VMware and Xen ?

This is a nice result in my case but from the Enterprise point of  
view, where VMware is more used than NetBSD/xen, this would prevent  
NetBSD from being chosen as a virtual machine :-(

Regards,
      Jo