Very bad read performance

[rahman]

Hi,

I am evaluating Proxmox 1.6 to migrate from ESXi 4.1. I installed it to a HP server with 4 core xenon and 4 GB ram raid 5 WITHOUT BBWC. I understand that without BBCW it is usual to have very poor writing performance. But I am getting 1-2 MB/s read speeds and the centos KVM vm runs very slugish. Is this expected? This will not be production server, our ESX servers runs on BBWC equiped servers and they have no IO problems. I will test proxmox on our new servers when we have them. But I just wanted to know if this poor io read is also related to BBWC or something wrong with my proxmox setup.


vz1:~# pveversion --verbose
pve-manager: 1.6-5 (pve-manager/1.6/5261)
running kernel: 2.6.32-4-pve
proxmox-ve-2.6.32: 1.6-25
pve-kernel-2.6.32-4-pve: 2.6.32-25
qemu-server: 1.1-22
pve-firmware: 1.0-9
libpve-storage-perl: 1.0-14
vncterm: 0.9-2
vzctl: 3.0.24-1pve4
vzdump: 1.2-8
vzprocps: 2.0.11-1dso2
vzquota: 3.0.11-1
pve-qemu-kvm: 0.12.5-2
ksm-control-daemon: 1.0-4



vz1:~# pveperf
CPU BOGOMIPS: 15999.96
REGEX/SECOND: 500187
HD SIZE: 57.09 GB (/dev/mapper/pve-root)
BUFFERED READS: 99.63 MB/sec
AVERAGE SEEK TIME: 9.32 ms
FSYNCS/SECOND: 33.26
DNS EXT: 135.55 ms
DNS INT: 1.12 ms (artvin.edu.tr)

 

[mmenaz]

Probably, writhe through seems to affect also read performance, that should be the only one tested by FSYNCS (don't know why).
Here are some data I've tested in the past:
p { margin-bottom: 0cm; } Proxmox 1.5 con kernel 2.6.24-11 su Xeon E5520 @ 2.27GHz
RAM 12 GB
HD 292.2 GB in RAID5 sas (3 units)
RAID bus controller: LSI Logic / Symbios Logic MegaRAID SAS 1078

p { margin-bottom: 0cm; } p { margin-bottom: 0cm; } [FONT=Courier New, monospace]CPU BOGOMIPS: 36266.74[/FONT]
[FONT=Courier New, monospace]REGEX/SECOND: 642373[/FONT]
[FONT=Courier New, monospace]HD SIZE: 66.93 GB (/dev/mapper/pve-root)[/FONT]
[FONT=Courier New, monospace]BUFFERED READS: 231.23 MB/sec[/FONT]
[FONT=Courier New, monospace]AVERAGE SEEK TIME: 4.53 ms[/FONT]
[FONT=Courier New, monospace]FSYNCS/SECOND: 159.03[/FONT]
[FONT=Courier New, monospace]DNS EXT: 131.92 ms[/FONT]
[FONT=Courier New, monospace]DNS INT: 111.74 ms (zzz.com)[/FONT]

Once enabled Writhe Through (I think you can enable it even without BBU, is only risky in case of failure/blackout, but since is a test machine... have a look at your controller setup):
p { margin-bottom: 0cm; } [FONT=Courier New, monospace]BUFFERED READS: 230.52 MB/sec[/FONT]
[FONT=Courier New, monospace]AVERAGE SEEK TIME: 4.53 ms[/FONT]
[FONT=Courier New, monospace]FSYNCS/SECOND: 3133.45[/FONT]

a HUGE improvement in [FONT=Courier New, monospace]FSYNCS/SECOND[/FONT].
In any case your simple "Buffered reads" is 2x slower than mine, and you don't have much ram either...

 

[rahman]

Thanks for the info, I will try it to see how it will perform. I as I said this is the only hardware that I have around and I know its not that powerfull. We have ordered 4 Fujitsu servers with all 2 cpu 8 core and two of them 72 gb ram, the others 24 GB each.

 

[udo]

Thanks for the info, I will try it to see how it will perform. I as I said this is the only hardware that I have around and I know its not that powerfull. We have ordered 4 Fujitsu servers with all 2 cpu 8 core and two of them 72 gb ram, the others 24 GB each.

Hi,
normaly more impact has the IO. Will say, use a good raid-controller with good disks (like 4*SAS-Disks as raid10) will be a huge different to slow controller which often sells with Servers.

Udo