Re: MC/S vs multiple sessions

I think what John and Julian were pointing out is that if the ordered or 
head of queue attributes are used along with simple attributes then using 
multiple sessions may not work properly. For the cases where MPIO is load 
balancing, the initiator will have to be sure it manually synchronizes the 
ordered and head of queue attributes with simple attributes. Also if the 
Queue Algorithm Modifier in the Control Mode Page is set to 0 then MPIO, 
doing writes to the same block over alternate sessions, will need to 
manually synchronize the writes.



I suspect that all MPIO writers are aware of these issues.


Eddy

----- Original Message ----- 
From: "Santosh Rao" <[email protected]>
To: "John Hufferd" <[email protected]>
Cc: <[email protected]>; "Qi, Yanling" <[email protected]>; "Santosh Rao" 
<[email protected]>
Sent: Friday, November 03, 2006 12:53 PM
Subject: RE: [Ips] MC/S vs multiple sessions


>
> I'm not aware of any need to ensure no I/O to the same LUN travels down
> another session (I-T-L nexus) while I/O is pending for the LUN on the
> first session (I-T nexus).
>
> Not sure which drivers are being referred to here when you say "most wedge
> drivers do not send I/O for the same LUN down different sessions (I-T
> nexi)."
>
> Most MPIO drivers I am aware of will all send I/Os down different
> I-T-L nexi of a lun to perform dynamic load balancing.
>
> Perhaps you were referring to I/Os to the same LBA range of a LUN.
> Serializing I/Os to the same LBA ranges of a LUN is the job of the upper
> layer application that owns the content residing on the lun such as a
> filesystem, database, etc.
>
> To answer Yanling's question, I think both solutions are equally feasible
> and the debate over MPIO vs. MC/S has been a long one in the past with
> marginal benefits from MC/S compared to utilizing existing MPIO
> infrastructure, avoiding command ordering bottlenecks, avoiding the
> creation of multiple layers performing very similar functionality.
>
> Thanks,
> Santosh
>
>
> On Wed, 1 Nov 2006, John Hufferd wrote:
>
>> Because of I-T-L Nexus issues, most wedge drivers (such as MPIO) do not
>> send I/O for the same LUN down different Sessions.  Even if they do,
>> they need to ensure that no I/O for the same LUN travels down another
>> Session while I/O for that same LUN is pending on the first session.
>>
>>
>>
>> With Multiple Connections per Session (MC/S), there are no such
>> restrictions.  Therefore the Initiator can send commands for the same
>> LUN down both connections at a rate that is only limited by the ability
>> of the storage controller to handle it.  It is also true that with MPIO
>> commands for other LUs can occur on the other Paths, but this may or may
>> not be the process which is most acceptable to the total environment.
>>
>>
>>
>> What this means is that performance is generally better with MC/S, but
>> the degree of performance improvement will depend on the design of the
>> Wedge Driver (MPIO) and the quantity of Multiple I/Os to the same LUN.
>>
>>
>>
>> .
>>
>> .
>>
>> .
>>
>> John L Hufferd
>>
>> Sr. Executive Director of Technology
>>
>> Brocade Communications Systems, Inc
>>
>> [email protected]
>>
>> Office Phone: (408) 333-5244; eFAX: (408) 904-4688
>>
>> Alt Office Phone: (408) 997-6136; Cell: (408) 627-9606
>>
>>
>>
>> ________________________________
>>
>> From: Qi, Yanling [mailto:[email protected]]
>> Sent: Wednesday, November 01, 2006 8:43 AM
>> To: [email protected]
>> Subject: [Ips] MC/S vs multiple sessions
>>
>>
>>
>> Assume my iSCSI target device has two network portals and my initiator
>> device has two network portals as well.  I could create two sessions or
>> create one session with two connections between the initiator and
>> target. I do have MPIO service on the initiator host that provides IO
>> load balance cross the two sessions. I could not see the differences
>> from the viewpoint of IO availability.  Which implementation is better
>> in term of IO performance?
>>
>>
>>
>> Thanks,
>>
>>
>>
>> Yanling
>>
>>
>>
>>
>>
>>
>>
>> Yanling Qi
>>
>> Engenio Storage Group - LSI Logic
>>
>> 512-794-3713 (Office)
>>
>> 512-794-3702 (Fax)
>>
>> [email protected]
>>
>>
>>
>>
>
>
> _______________________________________________
> Ips mailing list
> [email protected]
> https://www1.ietf.org/mailman/listinfo/ips 


_______________________________________________
Ips mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/ips