[ofa-general] Re: [Query] ib add path record cache

[Devesh Sharma]

Thanks for replying, My comments are as follows

On 5/11/07, sean.hefty <sean.hefty at intel.com> wrote:
> >One user command, reading path records from some file and passing this
> >to local_sa_cache module using standard entry point (read/write),
> >local_cache module is
> >assuming it as a incoming resolved path_record, and adding it to the
> >cache in normal fashion. possibly some device interface will be
> >required to be added.
> >port agent related issues needs to be looked into.
> >
> >Or some better idea if you have, can we discuss?
>
> This sounds fine.  I still just not understanding the reasoning behind

This can be treated as a facility similar to what we have in ARP table
for TCP/IP. Secondly this will help in debugging of some new up-coming
partially infiniband complaint hardware.

> populating the cache with dummy entries.  (I do think that being able to
> populate it from a file could be useful to initially load the cache in fairly
> static configurations.)
This is one more benefit we will get, It will prevent that initial
traffic generated by local_sa_cache module on the network, assume that
the cluster is big and every node is creating its cache DB, this will
generate a huge traffic burst, mutil-pathing will make the case even
worse. Generating a static path_record initially is a issue!
>
> I should note that the cache flushes old entries after it performs a full
> update.  So if these are entries that you want to remain in the cache after an
yes, I want them to remain in the DB, my idea is similar to the hard
coding of ARP table entries in TCP/IP.
How do you see this can be achieved?
> update, additional changes of the cache would be required.
>
> - Sean
>