SQL Server Performance Forum – Threads Archive
Another SQL Server on SAN QuestionI am planning to upgrade my SQL Servers and was wondering about a SAN configuration.
How does this sound: SAN = DS4300 w/Turbo
Servers – IBM x366 , 2 x 3.16 Intel Xeon, 8 GB (There are 6 of these)
SQL 2000 SAN consists of 2 EXP710 Drive enclosures with 14 73.4 GB,10k, 2Gb FC drives The servers are attached to the SAN via 2 Brocade H16 switches. I will have three Arrays. One for Data(7 drives (474GB), one for Logs (3 drives(203 GB)
and one for TempDB (3 drives(203 GB). All arrays will be Raid 10. Will this work? Thanks
there is not a yes or no answer, especially since you don’t say what your current memory config is and what your current disk load is.
btw, does each enclosure contain 14 drives? for a total of 28? the more important thing is to verify the performance characterstics on setting up this array, not waiting until after you go live, then, cannot change the configuration. for 14 drives in RAID 10 for the data, you should be able to get 1400 random read IOPS, 700 writes at low disk latency (10ms), assuming the data is distributed across the entire disk.
and over 2800 IOPS at high queue depth, high latency. two 2Gbit/sec FC port should be able to drive 330MB/sec sequential traffic, but i do not think the 7 or 14 data drives will sustain this number, for some reason, SAN systems do not do well in sequential ops. if you want to run both transactions and large reports concurrently, and both ops hit disks, then i recommend configuring the drive to sustain between 400-800MB/sec, which is more than what a SQL Server 2000 table scan will do, depending. 2 racks of 28 SCSI drives across 4 U320 channels can do this.
Each enclosure has 14 drives for a total of 28. So the arrays actually contain 14, 6, and 6 drives.
the reason it is very important to test your array is that for some strange reason, SAN vendors offer all these fancy "features" that have horrible performance consequences,
so many people endup spending a good chunk of money with crappy performance relative to the number of disks they purchased, and FC adapters
How would I go about testing the array? Any suggestions.
my recommendation is to use IOMeter (iometer.org, or search on source forge)
test random read and writes separately, random 8K, and sequential 1M.
for random read, test lowe and high queue depth other people like SQL IO Stress from the MS site (there are actually two versions of this)
there will also be a new disk stress from MS soon,
I have downloaded and installed the iometer program. I set it up using 15 workers. I selected the default access spec just to see it work. My question is how do I know when I am stressing the system? Response time, CPU?
i have some slides on this, will probably write an article later http://www.sql-server-performance.com/joe_chang.asp
http://www.sql-server-performance.com/Chang_SQL_Server_Storage_Performance.zip see also the top post in this section