Poor read performance, good write performance

sjors · Fri Jul 29, 2011 1:35 pm

I won't submit a crash report.
The software version I'm using is v5.6.1690.

I kept doing tests and I'm now getting numbers in the 910-930 range.
Please note that I'm doing the tests while one node is online.
I need to schedule downtime to perform clean tests.
This might also influence the numbers.

In the meantime, I did the same modifications on the inactive node and I'm now getting numbers in the same range.

I got one weird thing though, and I hope you can shed some light on this for me.
For one of the tests I'm using ATTO Disk Benchmark. (I think this is a great tool.)
This tool performs read and write tests on a disk using transfer sizes of 0.5KB to 8192KB.

The numbers I'm getting look good, as I reach transfer speeds of about 230MBps (on two interfaces,) but around the 1KB size, the write speed drops to about 11KBps and then rises again.
So, doing a write test of only 1KB size, it keeps giving me a low score.
Am I missing something here?

Thx,
George

Fri Jul 29, 2011 1:48 pm

V5.7 was recently released and we suggest you to try update path before proceeding with further experiments.

900+ is basically fine.

Yes you do miss a lot. Upgrade to V5.7 as it had numerous performance modifications. Update, re-run the test and share a screenshot with us and others here on forum.

sjors wrote:I won't submit a crash report.
The software version I'm using is v5.6.1690.

I kept doing tests and I'm now getting numbers in the 910-930 range.
Please note that I'm doing the tests while one node is online.
I need to schedule downtime to perform clean tests.
This might also influence the numbers.

In the meantime, I did the same modifications on the inactive node and I'm now getting numbers in the same range.

I got one weird thing though, and I hope you can shed some light on this for me.
For one of the tests I'm using ATTO Disk Benchmark. (I think this is a great tool.)
This tool performs read and write tests on a disk using transfer sizes of 0.5KB to 8192KB.

The numbers I'm getting look good, as I reach transfer speeds of about 230MBps (on two interfaces,) but around the 1KB size, the write speed drops to about 11KBps and then rises again.
So, doing a write test of only 1KB size, it keeps giving me a low score.
Am I missing something here?

Thx,
George

Fri Jul 29, 2011 1:55 pm

As I know you have RAID in you system. Could you kindly specify what type is it? 5, 6, 10?

sjors · Fri Jul 29, 2011 2:09 pm

I will schedule downtime to do the upgrade this weekend.

The system is using RAID-5.

When I'm done, I will post the screenshots.

George

Fri Jul 29, 2011 2:13 pm

Thank you. Keep us updated please.

Fri Jul 29, 2011 2:19 pm

RAID5 uncached is write pig. Make sure you run HA config with HUGE write back cache allocated.

sjors wrote:I will schedule downtime to do the upgrade this weekend.

The system is using RAID-5.

When I'm done, I will post the screenshots.

George

sjors · Fri Jul 29, 2011 2:53 pm

That's the thing.
I'm using a single server SAN.

I also upgraded from a lower version, so I cannot change cache settings.
(Or at least, I don't know how or where to do so.)

I also cannot remove my current volumes and recreate them.

Please advise.

George

Fri Jul 29, 2011 3:02 pm

Enable cache on RAID controller (assuming it has battery backup).

sjors wrote:That's the thing.
I'm using a single server SAN.

I also upgraded from a lower version, so I cannot change cache settings.
(Or at least, I don't know how or where to do so.)

I also cannot remove my current volumes and recreate them.

Please advise.

George

sjors · Fri Jul 29, 2011 3:21 pm

The system is using a PERC 5/E adapter, which has 256MB DDR2 memory for its cache.

It is enabled, but I don't know how much this 256MB will help.

And yes, it's battery-backed and the write policy is write-back.

Fri Jul 29, 2011 3:26 pm

Fine. Better then nothing but you'd better switch to RAID6 from RAID5 and use HA config with multi-gigabyte write back cache managed by StarWind.

sjors wrote:The system is using a PERC 5/E adapter, which has 256MB DDR2 memory for its cache.

It is enabled, but I don't know how much this 256MB will help.

And yes, it's battery-backed and the write policy is write-back.

sjors · Fri Jul 29, 2011 3:35 pm

The PERC 5/E doesn't support RAID6, so I won't be able to do that.

Besides that, how can I change the cache of existing volumes within Starwind?

Fri Jul 29, 2011 3:48 pm

Not with the version you have.

sjors wrote:The PERC 5/E doesn't support RAID6, so I won't be able to do that.

Besides that, how can I change the cache of existing volumes within Starwind?

sjors · Fri Jul 29, 2011 4:13 pm

With any other version?

Sun Jul 31, 2011 2:57 pm

V5.8 and up. The one with Cache Manager.

sjors wrote:With any other version?

CyberNBD · Sun Jul 31, 2011 6:31 pm

anton (staff) wrote:Yes. Both initiator and target sides please.

georgep wrote:i have the same problem with v5.7 Why do u say to put 2 instead of 1 ? 1 disables it right ?

During some tests in the past I found that disabling Delayed Ack only on the initiator side (using MS initiator) gave a slightly better performance than disabling it on both starwind and initiator machines. Did you compare?

sjors wrote:The system is using a PERC 5/E adapter, which has 256MB DDR2 memory for its cache.

It is enabled, but I don't know how much this 256MB will help.

And yes, it's battery-backed and the write policy is write-back.

How did you set disk cache policy? Using a similar MD1000 + Perc-5/E Setup I disabled it on the fast 15K SAS Spindles and enabled it on the slower 7.2K SATA spindles. This means the SAS array doesn't use the physical disk cache which gives better performance. For slower disks like the SATA's it's better to enable disk cache.