Testing Times 1

It's an interesting time with being asked to performance test/benchmark a new server setup for a migration of sql server and data centre.
Both current and new servers are SAN attached and it has been what is effectively the testing of the storage which has proved interesting.
 
To add to the mix I have a similar spec server with internal disk and I ran a couple of tests on one of my my home servers ( essentially a pc with lots of disks )
The three proper servers are modern top spec kit with similar processors - the standalone having the slowest clock speed
 
The hardware stacks up like this:
Server 1 - current:
x32 sql 2000 4 x quad xeon 2.9Mhz 32gb ram, san attached, raid 10 data ( 10 x 15k ), raid 1 logs ( 15k )
Server 2 - new:
x64 sql 2005 4 x quad xeon 2.9Mhz 64gb ram, virtualised san
Server 3 - standalone:
x64 sql2005 4 x quad xeon 2.4Mhz 32gb ram, raid 10 data ( 4 x 10k sas ), raid 1 logs ( 10k sas )
Server 4 - home:
x64 sql 2008 1 x dual core amd 4200 8gb ram, hardware raid 0 data ( 4 x 7.2k sata ), single 7.2k sata logs.
 
So on a basic test of inserting 1 million rows into a table ( has 4 secondary indexes and a text column ) generating 8.5 Gb of data where would you rank the  
servers?     Vote now on 087xxxxxx etc. etc.
 
Average results of several runs:- ( single table test )
 
Server 3     6 mins 30 secs            2,564 rows/sec   22Mb/sec  1,330Mb/min
Server 4     9 mins                    1,851 rows/sec   16Mb/sec     961Mb/min
Server 1     11 mins                   1,515 rows/sec   13Mb/sec     786Mb/min
Server 2     17 mins                     980 rows/sec   8.5Mb/sec    509Mb/min
 
( I'm currently putting a doc together on all the tests including the code )

Published 07 October 2008 15:41 by GrumpyOldDBA

Comments

# re: Testing Times 1

07 October 2008 16:26 by tonyrogerson

Why am I not suprised; congratulations on reproducing my results from the EMC rig I tested and its bad to see that SAN vendors haven't improved things in the past 2 years!

sqlblogcasts.com/.../1089.aspx

Its' absolutely terrible the arrogance of some of these SAN guys; the good SAN guys get it and listen and understand the problems with small IO's and SAN's in abilities to sustain a reasonable throughput at IO's of less than 64Kbytes (the transaction log max write size) and write thru (no caching).

Let me guess - these guys are blaming SQL Server for poor performance? Make sure you  use IOMeter to prove their rig doesn't perform well. There won't be much you can do about it though Colin - just highlight the problem and make sure its down as not a SQL problem (because look at IOMeter...)

Tony.

# re: Testing Times 1

09 October 2008 18:34 by royharv

Small point, but you will want to correct the CPU speeds from MHz to GHz.