I am sorry, i see iostat few other things, but what does that tell you about your IO.. Size, time, throughput.... bla I actually had to trace this crap down.. To better understand i mounted a drive directio To remove any buffer.. to track my IO i wrote a simple C program that did a write of various sizes.. using prex i followed the pid doing the writes... ----------------------------------------------------------------------------- prex -k enable io trace io ktrace on start C program or some IO... during that time collect pid and use truss to record writes... ktrace off ^d
tnfxtract io tnfdump io >> io.txt egrep <pid> io.txt |more ------------------------------------------------ by pid/device/buffer all that suff. here is what i saw.. if you follow the device # you will see that, depending on solaris config and FS your using, IO gets passed to the MD driver, then solaris, then to the SD driver.. all 3 layers should match.. so the md_maxphys maxphys all of the settings have to be set correctly or you get IO break down. and yes..I think there is actually 2 IOs seen 1 understood because of the sd/ssd driver shown for the 1 IO.. md device - > solaris driver -> sd/ssd driver config. If the md device is mirrored.. u see the write to both disk. do i have a life? Not really outside of golf. I had to chase this down becase on our EMC devices performance was awful .. found out that if you open up the IO at the veritas level, chop it at the solaris level and then chop it again you get better performance since EMC write cache performance degrades as IO is larger.. you also over run you IO_Subsystem and will see high IO waits and slow disk while everything looks normal... ODD!! yes... But watch out.. it will get you , and when/if it does.. it wll be a hard cookie to crack, unless you have seen it before. This message posted from opensolaris.org _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
