VM drive cache mode and performance

[giner]

Hello everybody,

Could anybody try to explain why cache=directsync and cache=none (which is write-back with host cache disabled) give completely same results? Shouldn't it be better on writing with cache=none?

------------------
Hardware:
- Dell PE 710
- PERC H700 Integrated
- RAID 0+1
- Current Cache Policy: WriteBack, ReadAdaptive, Direct, No Write Cache if Bad BBU
- BBU is in place

Proxmox:
- Proxmox VE 2.3 (is under some daily load during the tests)

Virtual machine:
- Windows 2008 R2 SP1
- 3 drives (virto HDD, 10GB raw image on ext3, i.e. on local storage)
-- drive E: cache=directsync
-- drive I: cache=none
-- drive F: cache=writethrough
------------------








Regards,
Stanislav

 

[spirit]

Hello everybody,

Could anybody try to explain why cache=directsync and cache=none (which is write-back with host cache disabled) give completely same results? Shouldn't it be better on writing with cache=none?

------------------
Hardware:
- Dell PE 710
- PERC H700 Integrated
- RAID 0+1
- Current Cache Policy: WriteBack, ReadAdaptive, Direct, No Write Cache if Bad BBU
- BBU is in place

Proxmox:
- Proxmox VE 2.3 (is under some daily load during the tests)

Virtual machine:
- Windows 2008 R2 SP1
- 3 drives (virto HDD, 10GB raw image on ext3, i.e. on local storage)
-- drive E: cache=directsync
-- drive I: cache=none
-- drive F: cache=writethrough
------------------








Regards,
Stanislav

I think it's because of your writeback cache of your perc HD700. try to put it in writethrough, and retry to do same benchs.

 

[giner]

> I think it's because of your writeback cache of your perc HD700. try to put it in writethrough, and retry to do same benchs.
But write-through would make it even slower, right?

 

[spirit]

> I think it's because of your writeback cache of your perc HD700. try to put it in writethrough, and retry to do same benchs.
But write-through would make it even slower, right?

Yes sure, it's just for bench directsync vs none.
(I think you should have lower performance with directsync and around same performance with none)