Disk performance

[hintalo68]

Hi Everybody

I have a debian 12 with 6.1.0-25 kernel. The hardware is HP dl380 g9 .
On the native linux kernel I have run this test three times.

sysbench fileio --file-total-size=5G prepare

The results was these.
917.16 MiB/sec
937.52 MiB/sec
934.25 MiB/sec

After that I have installed proxmox enviroment with 6.8.12-1-pve kernel, and run again the test.
The result was these.
621.06 MiB/sec
596.09 MiB/sec
594.96 MiB/sec

Why did the io speed drop so much , after installed proxmox ?
Thank

 

[waltar]

The hardware inside your HP dl380 g9 is what in # of disk(/s ?) and setup before and after install of pve ?

 

[floh8]

Pinne mal einen anderen kernel zum booten, der 6.8 is ein bisschen buggy.

 

[hintalo68]

Pinne mal einen anderen kernel zum booten, der 6.8 is ein bisschen buggy.

Hi

I have use non-sub distribution. deb http://download.proxmox.com/debian/pve bookworm pve-no-subscription
The lates kernel 6.8.12 :-( How can I upgrade ?

 

[hintalo68]

The hardware inside your HP dl380 g9 is what in # of disk(/s ?) and setup before and after install of pve ?

Smart Array Gen9 Controllers P440ar
5x 600GB SAS disk in raid 6 logical disk

 

[floh8]

Hi

I have use non-sub distribution. deb http://download.proxmox.com/debian/pve bookworm pve-no-subscription
The lates kernel 6.8.12 :-( How can I upgrade ?

Search for pin kernel Here in THe forum!

 

[waltar]

So just think about what you messured before ... you have 5 disk in raid6 which means 3 disks netto as 2 are (rotating) write parity. Measured >300MB/s per 600GB sas hdd is unrealistic and your results are cache results in case of even write and read unrealistic while the "slow ~600MB/s" looks same also.
Don't know the specs for a 600GB hdd yet but if I assume 133MB/s that max out at 400MB/s sustained and all other is caching mechanism.
That does not mean that pve has a more bad caching than your debian before but it still doesn't make it more easy to compare which circumstances were activ to get that different results on same hardware. I would not trust on sysbench and even the ever best benchmark is "the real running application" for the host !!