ProxMox Disk IO Perfrormance

[kylea]

Have been testing RHEL5.4 in a KVM versus a native install.

Both machines are identical

8GB RAM, 8 x 2.66 Xeon cores (2 x 4)

One has two 250GB Esata 3GB drives as single drives - 7200rpm

The other has a software mirror provided by RHEL so it sees a single 250GB disk. - 7200rpm

The Proxmox RHEL is reading at 114MB.sec

versus

49MB/sec for the native install

used hdparm to do the tests.

My Dell Laptop (E6500) with a 250GB Esata 7200RPM - reads a 89MB/sec

So its a very interesting performance difference.

Does Proxmox do some caching for the VM and will performance degrade as more data is added to the VM disk?

 

[dietmar]

The Proxmox RHEL is reading at 114MB.sec

versus

49MB/sec for the native install

KVM IO currently needs too much CPU (i guess your CPU is at 100% during tests?). So If you have
fast HDs the IO is limited by CPU power (IO uses only one core)

 

[tom]

take also a look on this post.
http://forum.proxmox.com/threads/3325-Performance-so-so?p=18768#post18768

 

[kylea]

I think you missed my point

The KVM RHEL instance is faster than the native RHEL instance

The questions are

1. Would the Software RAID make a difference?

2. Does the Debian server / Proxmox provide some additional cache support for the VM disk device?

 

[dietmar]

The KVM RHEL instance is faster than the native RHEL instance

Then you benchmark RAM/cache instead of disk IO.

1. Would the Software RAID make a difference?

It is an additional software layer, also using a cache.

2. Does the Debian server / Proxmox provide some additional cache support for the VM disk device?

yes, a 'write-through' cache.

 

[kylea]

Uhm - ok

I'll have a search for an IO testing tool

I used hdparm -tT <deviceid> of my tests. This just does a read test.

I will have access to a 4GB Fibre Channel array tomorrow - so I'll do some tests on that too.