Unsolved
This post is more than 5 years old
1 Message
0
524
March 14th, 2008 00:00
CX300 and MS Exchange performance issue.
Hi, all.
Well, we have run into a performance issue. The datastore of our Exchange 2003 has grown to some 180Gb. When the datastore grew, we initialle followed Microsofts recommendations and split the DB's onto two different volumes, but when we then later implemented Replication Manager, we had to consolidate back to one volume in order to get the setup qualified by EMC. Now comes the tricky bit: We only have room for one RAID1_0 RAID Group on the SAN with the current setup and as the Exchange had to share this space with another server, when the SAN was implemented, the Exhchange could not be allocated the necessary space as one coherent block, but was instead extended as a MetaLUN. Now some users claim that the Exchange has become really sluggish - but as not all users respond to problems like this, I am not 100% that it's general.
Now to my question: Is it to be expected that a RAID0_1 MetaLun performs so bad that the users will actually notice when the Exchange datastore reaches this size???
Well, we have run into a performance issue. The datastore of our Exchange 2003 has grown to some 180Gb. When the datastore grew, we initialle followed Microsofts recommendations and split the DB's onto two different volumes, but when we then later implemented Replication Manager, we had to consolidate back to one volume in order to get the setup qualified by EMC. Now comes the tricky bit: We only have room for one RAID1_0 RAID Group on the SAN with the current setup and as the Exchange had to share this space with another server, when the SAN was implemented, the Exhchange could not be allocated the necessary space as one coherent block, but was instead extended as a MetaLUN. Now some users claim that the Exchange has become really sluggish - but as not all users respond to problems like this, I am not 100% that it's general.
Now to my question: Is it to be expected that a RAID0_1 MetaLun performs so bad that the users will actually notice when the Exchange datastore reaches this size???
No Events found!
RRR
6 Operator
•
5.7K Posts
0
March 14th, 2008 02:00
However, when you use concattenated metas with luns on the same Raid Group performance is what you are experiencing: as long as you're on the fist lun, performance is equally distributed accross all physical disks. When your useage grows and you start using the 2nd lun in the meta (on the same Raid Group), imagine that the disk heads have to move back and forth between lun1 and lun2 and therefore adding extra latency in access times.
What you could do is temporarily migrate the metalun to some other Raid Group, then clean up some space on 2 or 3 Raid Groups, then add 2 or 3 equally large lun's on those Raid Groups and create 1 larme metalun as large as needed to "host" the data from the "old" meta you temporarily moved away. Move that "old" mettalun back to the new meta and performance should be better.
Also avoid using Raid Group which are already heavily used by other applications as 1 disk can only do a certain number of IOps (10k = 130 IOps, 15k = 180 IOps).
kelleg
6 Operator
•
4.5K Posts
0
March 19th, 2008 12:00
When you create a Raid Group, say a 2+2 R10 and you use four 10K fibre channel disks, the performance of the Raid Group will be determined by the combined performance of the two disk - about 120 IOPS per disk = 240 IOPS for the Raid Group.
If you have two LUNs in this RG and each LUN is receiving 200 IOPS, then the RG will be oversubscribed - 200 + 200 = 400 IOPS.
Get the "EMC CLARiiON Best Practices for Fibre Channel Storage" from PowerLink - just do a seach - there are a couple of diffenent vesions - pick the one that matches your flare version - also, if you have the older ATA disks, get release 24 as well, release 26 does not talk about the older ATA disks.
regards,
glen kelley
RRR
6 Operator
•
5.7K Posts
0
March 20th, 2008 08:00