Ben Rockwood wrote:
Rick McNeal wrote:I've not had a chance recently to run performance numbers using real disks as the backing store. Recently some hardware was made available to me, but I've not had a chance to run any performance testing because of other tasks. I'll see if I can make some time available tonight. Could you answer a couple of questions for me. (1) What size I/O are you doing to the target?8k up to 1MB.
When I use /tmp as the backing store the iSCSI target is able to run at wire speed once the packet sizes reach 8K.
(2) Are the builds from OpenSolaris? If so, I need to determine if these are debug builds or not. I've seen similar results when I run a machine with the debug bits.SX:CR. 49 on the target, 43 on the initiator.
The nightly builds are debug builds, but I believe the community builds are non-debug.
(3) Before your run did you happen to boot the machine using the -r option to boot. e.g. 'reboot -- -r' This too will kill performance.Nope. I'm curious why this would be though.
A -r causes all of the drivers to be loaded. At some point after boot the system will unload any drivers which aren't being used. I have found in the past that having all of these drivers loaded and then unloaded greatly impacts the performance.
(4) What benchmark are you using for iSCSI and NFS?dd and iozone. Although I'm open to suggestions on a better benchmark. I used bonnie++ from time to time.
I'm not familiar with iozone, I use an internal tool called vdbench. It doesn't really matter since I've been able to reproduce your performance problem and it's got my full attention.
Once I figure out what's going on I'll let you know. If this takes more than a couple of days I'll send out an update.
benr. _______________________________________________ storage-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/storage-discuss
-- ---- Rick McNeal A good friend will come and bail you out of jail...but, a true friend will be sitting next to you saying, "Damn...that was fun!" _______________________________________________ storage-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/storage-discuss
