How many sessions do you have writing to disk when you see decreased performance? Is it just one (meaning there would be 1 read and 1 write at the same time)?
Actually by think it again, if setup would be the same under Windows and UNIX then question is somehow pointless. However, I believe file system is not the same. I do not recall the details anymore, but once in the past we had an issue some config parameter for VxVM 3.5 used under Solaris 9, but can't remember the details now.
I guess if Windows is performing good with it and Solaris does not then it would not be CLARiiON either. Now the only question is how to isolate what is it. Of course, it this would be something between Solaris and CLARiiON I would assume EMC being aware of it and service request should be an option.
For all Performance Related issues, always check the "EMC CLARiiON Best Practices for Fibre Channel Storage - CLARiiON Release 22.pdf" available on Powerlink.
Check on page 36 - section called "Sizing ....".
Almost all performance issues on Clariion can be traced to:
1. insufficient spindles in the Raid Group - do you have enough spindles to handle the IO load you're sending to the LUN? For exmaple, a 4+1 Raid 5 using 10K FC disks can handle about 480 IOPS or about 40MB/sec bandwidth.
2. not enough memory allocated to write cache (leading to increased forced flushes) - allocate the maximum memory to write cache and the rest to read cache
The rest of performance issues are usually application related - i.e., SQL TempDB too small or located on the same physical disks as another application.
There are a lot of White Papers available on Powerlink that give good information about setting up the array for best performance (MetaLUNs, SQL, Oracle, Exchange, Clusters, etc).
Thanks for your answer. I think we followed all kinds of white papers and suggestions. This is used as an advanced file device type in NetWorker for backup to disk. In this case you can speak of bulk reads and writes. Only reads or only writes, no problem but one write and one read stream at the same time and performance/10 for both streams!! We planned to do some futher testing.
ble1
4 Operator
•
14.4K Posts
0
January 27th, 2007 11:00
Actually by think it again, if setup would be the same under Windows and UNIX then question is somehow pointless. However, I believe file system is not the same. I do not recall the details anymore, but once in the past we had an issue some config parameter for VxVM 3.5 used under Solaris 9, but can't remember the details now.
I guess if Windows is performing good with it and Solaris does not then it would not be CLARiiON either. Now the only question is how to isolate what is it. Of course, it this would be something between Solaris and CLARiiON I would assume EMC being aware of it and service request should be an option.
fvanwayenbergh
19 Posts
0
January 29th, 2007 01:00
Thanks
Felix
GaryPhipps
2 Posts
0
January 30th, 2007 03:00
could you let us know the outcome when you get it please?
Thanks,
Gary
kelleg
4 Operator
•
4.5K Posts
0
January 31st, 2007 07:00
Check on page 36 - section called "Sizing ....".
Almost all performance issues on Clariion can be traced to:
1. insufficient spindles in the Raid Group - do you have enough spindles to handle the IO load you're sending to the LUN? For exmaple, a 4+1 Raid 5 using 10K FC disks can handle about 480 IOPS or about 40MB/sec bandwidth.
2. not enough memory allocated to write cache (leading to increased forced flushes) - allocate the maximum memory to write cache and the rest to read cache
The rest of performance issues are usually application related - i.e., SQL TempDB too small or located on the same physical disks as another application.
There are a lot of White Papers available on Powerlink that give good information about setting up the array for best performance (MetaLUNs, SQL, Oracle, Exchange, Clusters, etc).
glenk
fvanwayenbergh
19 Posts
0
February 4th, 2007 23:00
Thanks for your answer. I think we followed all kinds of white papers and suggestions. This is used as an advanced file device type in NetWorker for backup to disk. In this case you can speak of bulk reads and writes. Only reads or only writes, no problem but one write and one read stream at the same time and performance/10 for both streams!!
We planned to do some futher testing.
Kind regards