vSphere 4.0 and drive performance
I added into my test ML 350 G5 a new HP Smart Array E200 controller and connected to it 2x SATA WD15EVDS (1.5TiB) drives in mirror. The VMFS partition was created via the VI Client. I have issue with the decreased write performance - 5-15MB/s average (few days ago I transferred 45GB image for two hours), reading average is 90+MB/s.
Is there a way how to solve this without recreating the VMFS partition?
Is there a way how to solve this without recreating the VMFS partition?
SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
jakethecatuk
Hey dlethe - if the OP doesn't have the cache module, they can't enable write cacheing - hence my question about the cahce module.
agreed, but checking to see of the module is installed won't solve the problem. They will still have to see if the write cache is enabled or disabled, so this saves them the step.
ASKER
Great, 44.1GB in 26 minutes. That should be an average of 28.2MB/s.
There is 128MB BBWC on both array controllers (built in E200i and second E200), FW version is 1.84. I used the HP SmartStart CD to start the ACU and enable the write cache. Read / write ratio is set to 50:50. I don't have the HPACUCLI installed on my vSphere server.
Is it possible to speed up the write even more?
There is 128MB BBWC on both array controllers (built in E200i and second E200), FW version is 1.84. I used the HP SmartStart CD to start the ACU and enable the write cache. Read / write ratio is set to 50:50. I don't have the HPACUCLI installed on my vSphere server.
Is it possible to speed up the write even more?
No - sorry.
As I said, the E200 are entry level cards and their performance is limited.
As I said, the E200 are entry level cards and their performance is limited.
That is about right, compared to your read performance. If you want to get more then it is going to be ugly. Changing NTFS cluster size and aligning the partition will make a nice bump, but it is a royal pain. Chances are your O/S is asking to read/write a whole 4KB at a time but the VM and RAID controller is bumping that up to at least 512KB at a time whether you like it or not.
Check this out for some nice tips that don't require totally rebuilding everything.
http://oreilly.com/pub/a/windows/2005/02/08/NTFS_Hacks.html
Check this out for some nice tips that don't require totally rebuilding everything.
http://oreilly.com/pub/a/windows/2005/02/08/NTFS_Hacks.html
ASKER
Thanks for the link, nice reading. But for now, the write speed is between two VMFS volumes (from 10k SAS drives to the SATA drives) using the VI Client.
Yes, but consider that the fastest write is the one that you don't have to perform. Disabling the last access time can be a biggie. I/Os on a VM are extremely expensive when it comes to overhead. Writing an additional 1MB every time somebody so much as looks at a file, or every time the o/s updates one of many log files adds up. I would do that now. You may not notice much on a specific benchmark, but overall the system will run a bit snappier just because you have fewer I/Os to perform overall every minute.
Do this, run perfmon and set it up to show you how many I/Os you run over a 5 minute period before and after disabling this. Don't kick off any benchmarks, let the system do whatever it ordinarily does during that time.
Do this, run perfmon and set it up to show you how many I/Os you run over a 5 minute period before and after disabling this. Don't kick off any benchmarks, let the system do whatever it ordinarily does during that time.