electric relaxation

…as promised. I used tbench, which is available in the Debian package archive:

$ sudo apt-get install dbench


My ‘test’ was to run a loopback tbench run with three clients. Make sure your VM isn’t doing anything intensive, then issue:

$ tbench_srv &
$ tbench 3


This will run a tbench test with three clients connected to the server, for 10 minutes with 2 minutes of warmup time. The comparison between VMI and non-VMI kernels on my machine was quite favorable: I got almost a 2X speedup when running with the VMI-enabled kernel! The numbers were 55MB/s before, and 95MB/s after.

I suspect this is some kind of pathological worst-case for the non-VMI kernel, since most benchmarks aren’t nearly this favorable. But nonetheless, I was happy with the result and so I didn’t question it too much.

More about dbench and tbench, from the manpage:

Netbench is a terrible benchmark, but it’s an “industry standard” and it’s what is used in the press to rate windows fileservers like Samba and WindowsNT. Given the requirements of running netbench (60 and 150 Windows PCs all on switched fast ethernet and a really grunty server, and some way to nurse all those machines along so they will run a very fussy benchmark suite without crashing), these programs were written to open up netbench to the masses.

Both dbench and tbench read a load description file called client.txt that was derived from a network sniffer dump of a real netbench run. client.txt is about 4MB and describes the 90 thousand operations that a netbench client does in a typical netbench run. They parse client.txt and use it to produce the same load without having to buy a huge lab.

dbench produces only the filesystem load. It does all the same IO calls that the smbd server in Samba would produce when confronted with a netbench run. It does no networking calls.

tbench produces only the TCP and process load. It does the same socket calls that smbd would do under a netbench load. It does no filesystem calls. The idea behind tbench is to eliminate smbd from the netbench test, as though the smbd code could be made infinitely fast.