[openib-general] Re: SDP close / linger problem

[Jerome Pioux]

I forgot: this is the way I run it:

On the receiver side:
LD_PRELOAD="/lib/libsdp.so" ttcp -r -p 5001 -l 1048576 -b 1048576 -f M -n 
5000

On the sender side:
LD_PRELOAD="/lib/libsdp.so" ttcp -t -p 5001 -l 1048576 -b 1048576 -f M -n 
5000 192.168.0.100

The linger is configured at 10s.
If you monitor the processes, you should see 10s delay for the sender 
process after all xfers were completed (or if you prefer, the receiver 
process should be gone 10s before the sender)

Jerome


----- Original Message ----- 
From: "Jerome Pioux" <jerome.pioux at bull.com>
To: "Michael S. Tsirkin" <mst at mellanox.co.il>
Cc: <openib-general at openib.org>
Sent: Tuesday, November 08, 2005 1:55 PM
Subject: [openib-general] Re: SDP close / linger problem


>I have attached ttcp.c
> I put the culprit code under #ifdef LINGER
>
> Thank you
> Jerome
>
>
> ----- Original Message ----- 
> From: "Michael S. Tsirkin" <mst at mellanox.co.il>
> To: "Jerome Pioux" <jerome.pioux at bull.com>
> Cc: <openib-general at openib.org>
> Sent: Tuesday, November 08, 2005 12:19 PM
> Subject: Re: SDP close / linger problem
>
>
>> Hmm.
>> Is the app small enough to post it here and use to reproduce the
>> problem?
>> If not, the traces with debug enabled might prove useful.
>>
>> Quoting Jerome Pioux <jerome.pioux at bull.com>:
>>> Subject: SDP close / linger problem
>>>
>>> Hi Michael,
>>>
>>> Not sure that my post earlier made it to the forum? ... so, I thought
>>> that I will send it directly to you since you are the SDP person.
>>>
>>> I have a problem at closure time with client-server apps when using SDP.
>>>
>>> The problem is that it ALWAYS takes the "LINGER" time value for one or
>>> the other side to complete the close() and regardless of what the value
>>> was set at (2s, 10s 30s, 60s...)
>>>
>>> >From my understanding, if the LINGER option is set, the close is queued
>>> >up on your send queue behind all other data potentially already queued
>>> >up at the time. But, if there are no data queued up, the close should 
>>> >be
>>> >immediate.
>>>
>>> On "my" app, this is always the receiver side that experiences the
>>> problem (the app is symetrical this is why there is a LINGER on the
>>> receiver side).
>>>
>>> I added a close in ttcp (ttcp does not use explicit close) with a linger
>>> time and the sender is now the one that always experiences this problem.
>>>
>>> I believe that, for both apps, all data have been sent (and received)
>>> correctly before the close - nothing (at least from the app view) is in
>>> the "pipe" (I had the app to report that before the closes).
>>>
>>> Finally, both app work fine using IPoIB - I meant that for the same
>>> tests, the closes are immediate, regardless of the LINGER values.
>>>
>>> Any idea please?
>>> I can provide traces if needed - please tell me what is needed and how 
>>> to
>>> get them.
>>>
>>> Thank you,
>>> Jerome
>>>
>>> ps: ia64 / RHEL4 / 2.6.12 / sn rev 3882
>>>
>>>
>>
>> -- 
>> MST
>


--------------------------------------------------------------------------------


> _______________________________________________
> openib-general mailing list
> openib-general at openib.org
> http://openib.org/mailman/listinfo/openib-general
>
> To unsubscribe, please visit 
> http://openib.org/mailman/listinfo/openib-general