Mike,

that is a very good point. I will adjust the session queue depth first
to give this a try.

Thanks,

Frank

On Tue, Aug 3, 2010 at 3:03 PM, Mike Christie <[email protected]> wrote:
> On 08/03/2010 09:12 AM, Frank Jansen wrote:
>>
>> Hi,
>>
>> During the testing that I am doing, I run into a bottleneck when I run
>> I/O to more than 1 device at the same time.
>>
>> In my current test configuration, I have the following setup:
>>
>> Target server has MegaRaid with 2 RAID 1E devices, which are mapped
>> through lvm into large and small LUNs. 4 LUNs are made available for
>> each RAID 1E devices through 2 instances of tgtd.  Small LUNs are on
>> control-port 0 and large LUNs on control-port 1, which are mapped to
>> separate nic ports (10GbE each).
>>
>> Initiator server had no problem discovery each of the portals and is
>> running I/O to 1 large LUN and 1 small LUN, which should separate all
>> traffic.  I'm using iozone to generate the traffic on both LUNs.
>>
>> What I am seeing is that I/O starts on 1 LUN and appears to wait until
>> the data file has been completely written and read on that LUN before
>> starting on the next.  Is it possible that this is really happening in
>> this manner?
>>
>
> It is possible. It depends on the order that the block and scsi layer sends
> the IO. If the device and session q depth is 16 and the block layer sends 16
> requests then it will all go to the one device and one device will sit idle
> until IO starts to complete. There scsi layer will check if devices could be
> starved and so when those commands complete it will make sure the next
> device is used first.
>
> You can either:
> 1. Make sure the device's queue depth is lower than the session q depth so
> that IO can be sent to all devices at the same time.
> 2. Do a target per lun.
>

-- 
You received this message because you are subscribed to the Google Groups 
"open-iscsi" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to 
[email protected].
For more options, visit this group at 
http://groups.google.com/group/open-iscsi?hl=en.

Reply via email to