I just ran navi analyzer because I wanted to get a measure of the IO on particular LUNs. There are not any performance issues but then I started looking at the figures for everything else and I saw a lot of disk crossing amongst other things. The array has never been configured to optimal performance so this is something Ive decided to start looking at. It may be that the users are used to this performance but I think it can be much better.
Im trying to understand why we have a lot of disk crossings if the alignment is correct, the IO is under 64KB and the element is set to 128. Its more for me learning to be able to size things correctly in future. To me this is sized perfectly therefore if there are disk crossings then im confused.
The hosts im seeing the issues with are all 2008 and LUNs presented to ESX4. The majority of VMs on the ESX hosts are 2008 too. I do use diskpart on the 2003 machines.
Yes. Element Size need to be considered before that one basic thing. Are there any Windows 2003 or lower servers attached to this clariion. If so, first try setting LUN Offset using diskpart command.
You might want to look at the 'File-system alignment' section of EMC CLARiiON Best Practices for Performance and Availability: Release 30.0 Firmware Update. This document is available on PowerLink.
Note that having a certain amount of drive crossings is 'normal' with larger I/Os. Verify that your I/O size is not larger than you think it is. In addition, MS Windows Server 2008 R2 automatically aligns LUNs.
Thanks for your response. My IO isnt any larger than 64KB and I have the default stripe element size. The majority of my IO is between 16KB-32KB and some times goes to 64KB.
Looking at that real quick. I would assume you have some other I/O not being monitored going on between 10:15:55 and 11:21:10. The I/O from 13:10:23 until the end is the expected behavior.
is the file system utilization over 80% on that particular Lun? you could try to defragment the filesystem with the appropriate defragmenation tool. if the filesystem is nearing full there may not be enough contiguous space to avoid disk crossing
Don't get "disk crossings" and "stripe crossings" confused - file-system alignment has more to do with "stripe-crossings" and has more affect on performance. Disk-crossings are somewhat normal and in most cases can be ignored. If you're running Windows 2008, the file-system alignment has already been assigned by Windows.
Better way to ask the question is are you seeing performance issues and if so which LUNs/Hosts, time frames when this occurs, then look at that LUNs in Analyzer.
orbital2
11 Posts
0
December 17th, 2010 03:00
I just ran navi analyzer because I wanted to get a measure of the IO on particular LUNs. There are not any performance issues but then I started looking at the figures for everything else and I saw a lot of disk crossing amongst other things. The array has never been configured to optimal performance so this is something Ive decided to start looking at. It may be that the users are used to this performance but I think it can be much better.
Im trying to understand why we have a lot of disk crossings if the alignment is correct, the IO is under 64KB and the element is set to 128. Its more for me learning to be able to size things correctly in future. To me this is sized perfectly therefore if there are disk crossings then im confused.
orbital2
11 Posts
0
December 17th, 2010 03:00
hi
The hosts im seeing the issues with are all 2008 and LUNs presented to ESX4. The majority of VMs on the ESX hosts are 2008 too. I do use diskpart on the 2003 machines.
Oli
mohas
1 Rookie
•
44 Posts
0
December 17th, 2010 03:00
Hi,
How did you confirm that the performance issue is because of disk crossing. Have you checked through Navisphere Analyzer?
mohas
1 Rookie
•
44 Posts
0
December 17th, 2010 03:00
Hi,
Yes. Element Size need to be considered before that one basic thing. Are there any Windows 2003 or lower servers attached to this clariion. If so, first try setting LUN Offset using diskpart command.
Thanks & Regards,
Mohideen
jps00
2 Intern
•
392 Posts
0
December 17th, 2010 08:00
You might want to look at the 'File-system alignment' section of EMC CLARiiON Best Practices for Performance and Availability: Release 30.0 Firmware Update. This document is available on PowerLink.
Note that having a certain amount of drive crossings is 'normal' with larger I/Os. Verify that your I/O size is not larger than you think it is. In addition, MS Windows Server 2008 R2 automatically aligns LUNs.
orbital2
11 Posts
0
December 17th, 2010 08:00
Hi
Thanks for your response. My IO isnt any larger than 64KB and I have the default stripe element size. The majority of my IO is between 16KB-32KB and some times goes to 64KB.
jps00
2 Intern
•
392 Posts
0
December 17th, 2010 11:00
Looking at that real quick. I would assume you have some other I/O not being monitored going on between 10:15:55 and 11:21:10. The I/O from 13:10:23 until the end is the expected behavior.
orbital2
11 Posts
0
December 17th, 2010 11:00
Here is the navi result. The write size is in fact under16KB
sidanat
51 Posts
0
December 20th, 2010 13:00
is the file system utilization over 80% on that particular Lun? you could try to defragment the filesystem with the appropriate defragmenation tool. if the filesystem is nearing full there may not be enough contiguous space to avoid disk crossing
kelleg
4 Operator
•
4.5K Posts
0
December 21st, 2010 13:00
Don't get "disk crossings" and "stripe crossings" confused - file-system alignment has more to do with "stripe-crossings" and has more affect on performance. Disk-crossings are somewhat normal and in most cases can be ignored. If you're running Windows 2008, the file-system alignment has already been assigned by Windows.
Better way to ask the question is are you seeing performance issues and if so which LUNs/Hosts, time frames when this occurs, then look at that LUNs in Analyzer.
glen