SQL Server Performance Forum – Threads Archive
Need super duper fast I/O solutionIf cost is NOT an issue, what would be the best solution out there that will provide the fastest disk I/O performance that one can get?
Don’t laugh. I’m being serious here.
In the early phase of research to see what’s out there. We have a few modeling apps that crunches through a ton of historical data stored either SQL or Oracle DBs.
I’m assuming that this will be a high end SAN solution of some sort instead of DAS.
We’ve been using an IBM DS4300 as an entry level into SAN and haven’t been too impressed at all.
It looks like we might have to step up to something like an EMC Clariion or Symmetrix SAN line. Kaching!
If you have any good links for some SAN performance info (non vendor centric), that would be great also. Why is it so hard to find non vender specific performance data?!
Just throwing this out hoping to get some directions to start.
did you want:
or 2) high cost? if 2) then go with the SAN
if 1) then go with direct attach storage
the concept is shortest path to metal (or iron oxide in this case) if price really is no object, then look at the TMS ram-san stuff. but if you understood how SQL Server uses disks, you will realize most tasks can be handled with disk drives, a few specialties are suitable for solid state storage
We want PERFORMANCE.
I’m surprised with your opinion actually but saving $$$ is always good. Ultimately, we need a SAN also to take advantages of it’s mirroring, SAN copy capabilities. We’re hoping it’ll be fast also to attach our modeling servers to it also. If a DAS solution for these modeling servers, then so be it. As for a DAS solution, what would we need to get the fastest disk I/O possible? 2-3 RAID controllers
4-8 DASD with 14 HDs each. What would be the max before we saturate the motherboard bus basically?
Is there a usuable "volume manager" solution for DAS to stripe data across arrays for increased I/O performance?
you need to determine what performance level you want on the 2 characteristics:
1) random IOPS on a data set of X GB (important, otherwise someone will just do a test to cache)
2) sequential bandwidth set goals for the above, then get a quote from a SAN vendor, and ask them to demonstrate it using SQL Server to drive the load.
On the random IOPS, if you need 40K IOPS on a database of 1TB, be sure to test with random IOPS to 1TB, otherwise they demostrate random IOPS to the 4GB cache, which is totally stupid. most systems can handle any random IOPS requirement if you can connect enough disks for sequential: MS has demonstrated 2.4GB/sec on the Opteron platform,
the old Xeon (533MHz FSB) could do 1.3GB/sec
the new Xeon platforms (800MHz FSB, PCI-express) can do atleast 1.7GB/sec, probably more (limited by the disks I had for this test). If these systems cannot do the same 2.4GB/sec as the Opteron, I would post this info along with test details
In the past, Intel did not show proper appreciation for the importance of sequential bandwidth in their chipsets the HP Itanium platforms can do atleast 4GB/sec (rx8640) and 12GB/sec (Superdome)
I just read your article "System and Storage Configuration for SQL Server".
Great article as I got some good info from it to start with. For consistent comparisons, are there good tools to gather IOPS stats?
Would Microsoft SQLIO be ok?
for now use IOMeter,
my issue with SQLIOStress is the limited file size 3GB?
there are actually 2 tools from MS, can never keep the names straight some time later, i will release some SQL scripts for generating IO load