Linux Containers - benchmarks

Results

Testing Tools	r3 nodes: x86_64, 2cpus+HT, config	r1 nodes: i686, 2cpus+HT	...
Dbench	Result	To be done	...
Tbench	Result	To be done	...
Kernbench	Result	To be done	...

TODO

Homogeneous

From Kirill: All binaries should be built statically to work the same way inside host/guest or you need to make sure that you have exactly the same versions of glibc and other system libraries. At least glibc can affect performance very much :/

For the moment nodes are using a RHEL4U3 installation. As all virtual guest are really easy to set up with a Debian/Sarge distro (thank to debootstrap), I will switch our nodes to Sarge. Like this, there will be no glibc (or whatever) package difference between nodes.

File system bench

From Kirill: can you please remake this test for OpenVZ with iosched=as?

I took a look at Documentation/block/as-iosched.txt (under a kernel source tree) and it seems that we can easily change linux io scheduler. Just need to append to the boot commmand line:

elevator=as

Oprofile

We will need to add Oprofile to all tests in order to provide clues about performance differences between the container solutions.

Scalability

Some interesting measures could be done with concurrent benchs on several guest. For example in order to see the cost of splitting node resources.

Misc.

Last update 2006 july 10. Maintained by Calmels Clement