[openib-general] Re: [openib-commits] r2695 - gen2/users/jlentini/linux-kernel/dat-provider
James Lentini
jlentini at netapp.com
Fri Jun 24 07:42:03 PDT 2005
On Fri, 24 Jun 2005, Hal Rosenstock wrote:
> On Thu, 2005-06-23 at 15:18, jlentini at openib.org wrote:
>> Author: jlentini
>> Date: 2005-06-23 12:18:14 -0700 (Thu, 23 Jun 2005)
>> New Revision: 2695
>>
>> Modified:
>> gen2/users/jlentini/linux-kernel/dat-provider/dapl_openib_cm.c
>> Log:
>> Retry normal packets but not the CM messages. CM message retries are
>> linked to dat_ep_connect()'s timeout value
>
>> Signed-off-by: James Lentini <jlentini at netapp.com>
>>
>>
>> Modified: gen2/users/jlentini/linux-kernel/dat-provider/dapl_openib_cm.c
>> ===================================================================
>> -- gen2/users/jlentini/linux-kernel/dat-provider/dapl_openib_cm.c 2005-06-23 19:15:12 UTC (rev 2694)
>> +++ gen2/users/jlentini/linux-kernel/dat-provider/dapl_openib_cm.c 2005-06-23 19:18:14 UTC (rev 2695)
>> @@ -39,10 +39,10 @@
> ...
>> #define DAPL_IB_CM_RESPONSE_TIMEOUT 20 /* 4 sec */
>> -#define DAPL_IB_MAX_CM_RETRIES 4
>> +#define DAPL_IB_MAX_CM_RETRIES 0
>
> Until the algorithm previously discussed is implemented, changing
> DAPL_IB_MAX_CM_RETRIES to 0 makes this more frail (in the case of a lost
> REQ, etc.).
I made this change because this is what we intended to do in the
original "timeout implementation" change. I've added an item to the
TODO list to account for address resolution time and CM retries in
the timeout value.
In my testing, the lack of retries hasn't been an issue. If it is, we
can boost it up and account for it in the timeout value.
More information about the general
mailing list