May 14, 2009 at 1:58 am
Dear Forum,
a few days ago we got a brand new EMC AX4 SAN System and a new HP DL360G5 server, which is supposed to become our new SQL Server. However, something doesn't seem right. The best configuration so far yields a write-performance of 45 MByte/sec. I have no experience with Raid-5, but this is by far less than I expected.
What would you consider "reasonable" transfer rates for a regular "big-file copy" ?
Here is the details:
I copy a big file from drive E to drive F, that is from one LUN (on SP-Controller A) to another LUN (on SP-Controller B). Both drives are NTFS formatted with 64K Blocksize and 64K alignment(used diskpart). Either LUN is set up as a RAID-5 array with 4+1 Disks. All Disks are SAS 147Gig 15KRPM. Write Cache is active (360 MByte), Read-Cache is active (75MByte). The AX4 SAN is connected using two fiber-channel HBAs, each running 4GBit/sec. There's no online virus-scanner.
Without the built-in SAN write-Cache, the performance degrades to constant 9 MByte/sek. That is nine Megabytes per second ! On a brand new SAN from EMC.
Something must be wrong. Any ideas ?
Regards
Kay
September 2, 2009 at 3:50 pm
I hope this isn't a stupid question but have you updated all drivers for controller and raid. firmware too?
i remember once i was in the same situation and after a phone call to the vendor, all it took was a firmware upgrade to my controller card.
September 3, 2009 at 7:14 am
September 3, 2009 at 8:34 am
The performance of SAN depends on so many things that you really need to test it to make sure you have the expected performance. The 9 MB/Sec you mentioned is well below what I would expect, so you need to get it fixed before going live.
On local attached storage using RAID-5 with 6-10 disks, I usually see around 600 MB/sec for sequential 64K read, 300 MB/sec for sequential 64K writes, 200 MB/Sec for random 64K reads, and 60 MB/Sec for random 64K writes using the testing method described below.
I prefer to do IO testing using the SQLIO utility with a test file size of at least 8 GB (to overcome the effect of cache) using 16 threads with 8 outstanding requests, and run tests of sequential/random and read/write for block sizes of 4K, 8K, 16K, 32K, 64K, 128K, and 256K that run at least 1 minute each. That should give you a good idea of the IO performance you can expect. I run a test like this on every new server, and it has been very effective at identifying IO problems before they get to production.
September 3, 2009 at 12:47 pm
9 Mb/second = 540 Mb/minute ... Hmmm ...
In theory, dual channel, dual path 4 Gb fiber cards, puts you at a theoretical maximum of a about 1.8 Gb/minute thoughput each way if your SAN read/write cache are both large enough and balanced properly.
In order to achieve that there has to be sufficient cache on the SAN and ample spindles for the SAN LUNs and potentially Meta-LUNs on this EMC SAN. Unless you are doing thousand to tens of thousands of SQL transactions per second RAID 5 and RAID 10 stack up reasonable even.
From what I read about your configuration with the post you are doing OK - but it can be better !
RegardsRudy KomacsarSenior Database Administrator"Ave Caesar! - Morituri te salutamus."
Viewing 5 posts - 1 through 4 (of 4 total)
You must be logged in to reply to this topic. Login to reply