[openib-general] opensm RFC: access switch object via node

[Eitan Zahavi]

Sasha Khapyorsky wrote:
> On 14:18 Mon 27 Nov     , Eitan Zahavi wrote:
>   
>> Hi Sasha,
>>
>> The idea is to add a check to drop manager such that the database is
>> consistent:
>> 1. A node has a switch pointer but the map of switch by guid is missing
>> that pointer 
>> 2. The switch by guid table has entries that do not have a node (I think
>> this is already covered)
>>     
>
> I see. Probably this is "overkill" in general, but it could be not bad
> to have this some period for debug purposes. I will think about this
> option. Thanks for idea.
>
> Sasha
>
>   
The main reason to have drop manager is to handle inconsistencies caused 
by dropped packets in the discovery process.
E.g a node that is marked as a switch but the SwitchInfo SubnGetResp 
never made it to the SM.
Maybe this can not happen in the case of the new pointer. But I think 
writing a standalone checker for the DB
consistency is a good debugging method.
>> Eitan Zahavi
>> Senior Engineering Director, Software Architect
>> Mellanox Technologies LTD
>> Tel:+972-4-9097208
>> Fax:+972-4-9593245
>> P.O. Box 586 Yokneam 20692 ISRAEL
>>
>>
>>     
>>> -----Original Message-----
>>> From: Sasha Khapyorsky [mailto:sashak at voltaire.com]
>>> Sent: Monday, November 27, 2006 2:21 PM
>>> To: Eitan Zahavi
>>> Cc: Hal Rosenstock; openib-general at openib.org
>>> Subject: Re: [openib-general] opensm RFC: access switch object via
>>>       
>> node
>>     
>>> On 13:40 Mon 27 Nov     , Eitan Zahavi wrote:
>>>       
>>>> Looks like a good idea to me.
>>>>         
>>> Glad to hear this.
>>>
>>>       
>>>> I propose adding this rule to osm_drop_mgr.c too so cases where this
>>>> rule breaks are caught.
>>>>         
>>> Do you mean (node->sw == NULL) checks? If so, I agree - good idea.
>>>
>>> Sasha
>>>
>>>       
>>>> Thanks
>>>>
>>>> EZ
>>>> Sasha Khapyorsky wrote:
>>>>         
>>>>> Hi,
>>>>>
>>>>> I noticed that in most of cases when switch object lookup is
>>>>> performed in OpenSM it is done by searching sw_guid_tbl map by
>>>>>           
>> switch
>>     
>>>>> node's GUID value.
>>>>>
>>>>> For me it looks better and simpler just to keep pointer to switch
>>>>> object as member of its node structure. So for switch nodes such
>>>>> pointer will point to node's valid switch object and for other
>>>>>           
>> nodes
>>     
>>>>> this will be initialized as NULL.
>>>>>
>>>>> Then instead of sw_guid_tbl map searching we will have switch
>>>>>           
>> object
>>     
>>>>> pointer directly available from the node structure (and sw_guid_tbl
>>>>> map can be replaced by "cheaper" linked list).
>>>>>
>>>>> Other positive "side effect" is nice way (one more) to detect
>>>>> switch/non-switch node type just as (node->sw != NULL).
>>>>>
>>>>> Any comments?
>>>>>
>>>>> If no real objections I can prepare patches.
>>>>>
>>>>> Sasha
>>>>>
>>>>> _______________________________________________
>>>>> 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
>>>>>
>>>>>           
>
> _______________________________________________
> 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
>