> Unless mistaken, the time to execute 1 IO is: (122120-3700)/4722 = 25ms
> so 25ms for 1 IO seems very slow for me. Am I right?
Looks correct to me- its seems very slow.
I've got a SAN here doing 12ms and I consider that to be slow. 25ms is awful.
I consider reading a page from a SAN in 2ms to be good. I have servers which will sometimes do 0.5ms which is fast.
Either
- your devices aren't set up correctly ?
- your SAN is slow or overloaded ? (What SAN are you using ? EMC? HDS ? etc and what model ?)
- your Sybase server is busy ? (Were you the only user on the Sybase server ?)
Not sure I can help much with these but here's so thoughts.
First - check your devices are set up correctly.
If you've got SAN (which it looks like you have) then contact the SAN team. Ask them what *YOUR* throughput is.
In my experience, you'll get ignored and brushed off. They'll talk about "IOPS" - ignore this - its useful for SAN performance but has no relationship to your individual database performance. Ask them repeatedly the worse case time for a page read. Get the SA's to help with showing the IO waits. SAN teams will usually only talk to SA's.
Ask them what page size the SAN uses. (We've gained a lot of performance by increasing the Sybase page size to match the page size of the SAN. Our SAN is 16k pages, so 16k Sybase pages is good - generally not much to be gained to be larger than the SAN page size. It took a while to rebuild but performance is better!)
What size are you using ?
Askt hem if the SAN does dynamic tiering ? Heavily used pages are held "higher" up the SAN.
When we installed a new server on the SAN - performance was initially very poor as we were automatically allocated disks lower down the SAN. As we used the server, the SAN automagically moves the heavily used pages to higher up the SAN. Now when I install a new server, I ran "unbindcache, select count(*) from ...." repeatedly for about 24 hours - this will cause the SAN to move those tables to better disks (obviously it means other teams move down the SAN. In fact, you can probably see the dynamic-tiering by repeatedly running the select count(*), clearing sybase cache and checking the times. In my experience the 3rd run is twice the speed of the first run.