February 28, 2005 at 6:51 am
Hi All,
Has anyone used this and have any comments?
I am having a problem where some machines we ship perform very differant to others.
Same or near similar spec drives etc and some are just dog slow while others fly.
They are all built by the same guy with varying levels of hangover.
There is an extract that I am curious about.
"Finally, it was found that a well known manufacture had a firmware bug that lead to blue screens when outstanding I/O levels exceeded a certain level of outstanding I/O requests and depleted non-page pool."
Anyone know of the vendor? We have the occasional blue screen for no reason. Sometimes it's when someone is connected via pcanywhere, sometimes just for the fun of it.
I am also looking for a utill that will benchmark SQL for me and give me a score. All machines have to equal or better this benchmark. But that's another topic.
I would just like to get some info / stories about SQLIOStress if there are any.
Cheers,
Crispin
Cheers,CrispinI can't die, there are too many people who still have to meet me!It's not a bug, SQL just misunderstood me!
February 28, 2005 at 11:36 am
I have used SQLIOStress many times to prove hardware configuration problems. What specifically to you want to know ?
Regarding "Same or near similar spec drives etc and some are just dog slow while others fly." , check the following:
The Network Interface Card duplex and speed: under high network loads, when duplex and speed are set to "auto", the information exchange between the network hardware (such as the routers) to handshake the appropriate duplex and speed can get confused and have an adverse affect on thruput. Set the duplex and speed manually to the full capabilities of the the NIC.
For Windows performance configuration, which are slightly different between 2000 and 2003, set to optimize for background programs e.g. the SQL Server service.
For servers dedicated to SQL Server, consider turning off file and print services for the NICs.
For servers with large amounts of memory, check the boot.ini to see if the /3gb or /pae are configured differently.
SQL = Scarcely Qualifies as a Language
February 28, 2005 at 11:54 am
Hi Carl,
The problems (most of the time) are not network related. When I run the app local to the machine, it's as slow as when run from the network so I recon I could rule that out.
It has something to do with the disks, sometimes a query can take ages to run while other times (machines) it's almost instant.
re: SQLIOStress.
Running the defaults ok? Was going to run it with the data file on drive 1 and log on drive 2 as per SQL setup. In the documentation it states run one, wait a while then kick off another etc until you have 6 running or hitting around 60% on your machine. Seeing as the default will run just shy of 20 hours, I would hate to waste 20+ hours on a dud test. Any other nice switches to hand it?
I am currently running it with:
SqlIoStress.exe -FD:\Temp\Stress1.mdf -LD:\Temp\Stress1.ldf -C -S4090 -N -H -P
as those are the ones I think are most important to me now.
Cheers,CrispinI can't die, there are too many people who still have to meet me!It's not a bug, SQL just misunderstood me!
February 28, 2005 at 1:17 pm
Regarding the differences between the same query having major differences in response times on different servers, this could be a problem with the physical database such as out of date statistics, differences in data, fragmentation of indicies and so forth.
To eliminate these variables, recommend you copy a database from a server with good performance to a server having poor performance.
Instead of SQLIOStress.exe, which does have a long run time, you could use the SQLIO Benchmark instead which runs in much less time. See http://www.microsoft.com/downloads/details.aspx?FamilyId=9A8B005B-84E4-4F24-8D65-CB53442D9E19&displaylang=en
From your posted SqlIoStress.exe parameters, I noticed that the data file and log file are specified to be on the same partition, which will have a major effect on the run time Here are some benchmarks that I recall:
A configuration where the Data is on a dedicated RAID 1 pair of drives and the log is on a separate and dedicated RAID 1 pair of drives versus:
Data and Log on the same set of Raid 5 drives results in a 12 times longer run time.
Data and Log on the same pair of Raid 1 drives results in a 6 times longer run time.
SQL = Scarcely Qualifies as a Language
Viewing 4 posts - 1 through 3 (of 3 total)
You must be logged in to reply to this topic. Login to reply