[ofa-general] [PATCH 0/6 v2] opensm: Unicast Routing Cache

[Yevgeny Kliteynik]

Hi Sasha,

I'm sending v2 of the patches:

  - patch 1/6: move lft_buf from ucast_mgr to osm_switch
    No change whatsoever, just rebased to the new master

  - patch 2/6: Add "-A" or "--ucast_cache" option to opensm
    No functional change, changed appearance of the help message
    to match the recent OpenSM help changes.

  - patch 3/6: adding osm_ucast_cache.{c,h} files.
     + All your patches integrated, many small fixes, including
       all the fixes from the review.
     + Port array allocation reworked to save many reallocations
     + Better malloc() handling: checking for returning NULL,
       no casting after malloc().
     + Back-2-back connections - cache is now disabled when there
       are no switches, but it also takes care of the case when SM
       port is disconnected and then connected with back-2-back link.
       Besides adding check for zero switches in fabric, I found only
       one back-2-back connection check that could be removed - in
       function osm_ucast_cache_add_node().

  - patch 4/6: adding new cache files to makefile.
    No changes, just rebased.

  - patch 5/6: integrating unicast cache into the discovery
    and ucast manager.
    This patch includes your changes (having ucast_cache_process
    function that does all the job instead of integrating into
    osm_ucast_mgr_process). Also, the cache is now a part of the
    ucast manager struct, which simplifies a code and saves some
    functions.

  - patch 6/6: man entry for cached routing.
    No changes, just rebased.


The job that still needs to be done:
  - Check how the cache handles port moving during discovery.
    Might be a bug there.
  - Check how unicast manager handles fast reset of switches.
    AFAIK SM will now write the LFT there - need to fix it
    (unrelated to cache, general ucast mgr issue)
  - Optimize LFT usage - simplify current switch LFT,
    hold two LFTs (current and cached) only when these LFTs
    are not identical.

-- Yevgeny

Yevgeny Kliteynik wrote:
> Hi Sasha,
> 
> Sasha Khapyorsky wrote:
>> Hi Yevgeny,
>>
>> On 03:29 Fri 10 Oct     , Yevgeny Kliteynik wrote:
>>> Thanks for the review and the patches. Didn't manage to address
>>> all your comments yet - will do it tomorrow.
>>> One question though: how to deal with the incremental patches that
>>> you sent me? Should I apply them to my branch and then issue one
>>> V2 patch instead of the old one, or will you apply the original
>>> patch, followed by all the incremental (yours and mine)?
>>
>> It is up to you. You can merge all in single V2 (guess it is simpler)
> 
> I'll send a v2 patches for 3/6 and 5/6 when I'm done fixing all the stuff.
> 
> -- Yevgeny
> 
>> or leave it unchanged and I will apply later. Except integration patch
>> others are not critical IMHO.
>>
>> Sasha
>>
> 
> _______________________________________________
> general mailing list
> general at lists.openfabrics.org
> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
> 
> To unsubscribe, please visit 
> http://openib.org/mailman/listinfo/openib-general
>