Erez Zilber wrote:
> Mike,
> 
> I'm still investigating the performance problems in open-iscsi over iSER, and 
> I think that I see the cause of the problem:
> 
> I've added an outstanding commands counter. It's incremented when 
> iscsi_queuecommand is called and decremented when iscsi_complete_command is 
> called. After decrementing the counter, I save it's value in some global 
> array. I print this array when iscsi_session_teardown is called.
> 
>>From what I see, in ~25% of command completions, the command queue is empty 
>>(i.e. outstanding_cmds = 0). This can explain the low throughput that I see 
>>with iSER.
> 
> What could be the cause for that? Is there anything not well configured in 
> the scsi_host_template that we use? Anything else?
> 

I saw this when using all io schedulers except noop. When I set the 
queue to use noop, The iscsi queue was full (host_busy was greater than 
1) most of the time and thoughout was higher. I was going by just checking

/sys/class/scsi_host/hostX/host_busy

which if I underatnd your patch right is similar to what you are doing.

I will run your patch to double check though.

Also I think you can tweak the other schedulers to fix the problem, but 
I have not found the right combo yet.

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups 
"open-iscsi" group.
To post to this group, send email to open-iscsi@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at http://groups.google.com/group/open-iscsi
-~----------~----~----~----~------~----~------~--~---

Reply via email to