[PVFS-developers] "Parallelizing" manager cross-talk to IODs

[White Stuart - stwhit]

D'oh!

See attached patch.

-----Original Message-----
From: Rob Ross [mailto:rross at mcs.anl.gov] 
Sent: Wednesday, April 28, 2004 3:43 PM
To: White Stuart - stwhit
Cc: 'pvfs-developers at www.beowulf-underground.org'
Subject: Re: [PVFS-developers] "Parallelizing" manager cross-talk to IODs


Hi Stuart,

I think that this would be a significant improvement to the mgr.  Send 
along the patch and we'll have a look at it.

Thanks,

Rob

On Wed, 28 Apr 2004, White Stuart - stwhit wrote:

> Hello all,
> 
> The pvfs manager frequently sends requests to each of the IODs in 
> order to process requests.  It calls send_req() to accomplish this.  
> The send_req() routine contains a loop that loops through all the 
> IODs, sending the request and waiting for the ACK from the IOD.  To 
> illustrate:
> 
> forall (IODs)
> {
>    send_request_to_IOD();
>    wait_for_ACK();
> }
> 
> What results from this is that the IOD requests are sent in a very 
> sequential manner; the manager waits for an ACK from an IOD before 
> sending the request to the next IOD.  In general, this is acceptable.  
> However, in cases where the IOD takes a significant amount of time to 
> service the request, this method of sending requests to the IODs is 
> inefficient.  The time it takes for the manager to service the request 
> is the sum of the times it takes for the IODs to service their 
> requests.
> 
> I would like to suggest that we restructure the code to look like 
> this:
> 
> forall (IODs)
> {
>    send_request_to_IOD();
> }
> 
> forall (IODs)
> {
>    wait_for_ACK();
> }
> 
> This effectively "parallelizes" the requests to the IODs.  Now, the 
> time it takes the manager to service the request is only equal to the 
> longest time one IOD takes to service its request.
> 
> I've attached a patch against the 1.6.2 source tarball that implements 
> this change.
> 
> I ask that the PVFS developers take a look at this suggestion and see 
> if it makes sense to them.  This change *seems* to make sense to me, 
> but perhaps I'm overlooking something.  All feedback is greatly 
> appreciated!
> 
> Stuart
> 
> 
> **********************************************************************
> The information contained in this communication is confidential, is 
> intended only for the use of the recipient named above, and may be 
> legally privileged. If the reader of this message is not the intended
> recipient, you are hereby notified that any dissemination, 
> distribution, or copying of this communication is strictly
> prohibited.
> If you have received this communication in error,
> please re-send this communication to the sender and
> delete the original message or any copy of it from your
> computer system. Thank You.
> 
> _______________________________________________
> PVFS-developers mailing list 
> PVFS-developers at www.beowulf-underground.org
> http://www.beowulf-underground.org/mailman/listinfo/pvfs-developers
> 
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: pvfs-mgr-parallel-crosstalk.patch
Type: application/octet-stream
Size: 738 bytes
Desc: not available
Url : http://www.beowulf-underground.org/pipermail/pvfs-developers/attachments/20040428/69974ed8/pvfs-mgr-parallel-crosstalk-0001.obj