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

[Devesh Sharma]

On 5/29/07, Sean Hefty <sean.hefty at intel.com> wrote:
> >Ok, but, by that time we can keep the framework ready?
>
> I plan on re-submitting the cache for 2.6.23.  Beyond that I won't have the time
> to work on enhancements for a few weeks.  I will happily review any patch
> submissions though.
Ok, Soon I will post a patch related to this.
How static PR file will be generated? Needs to be discussed.
>
> >How this will be managed? This will add extra startup time in the
> >cluster, because cluster will be usable only after last cache has been
> >enabled. Am I right?
>
> I would word this differently: we can improve the time required to load the
> cache, versus stating that the cache adds extra startup time.
>
> The cache is not necessary to use the cluster, so doesn't force extra startup
> time.  Cache misses would simply be forwarded directly to the SA.  If the first
> application to run on the cluster isn't establishing all-to-all communication
> between the nodes then there may not be any reason to delay starting the app.
>
> Even if the first app does establish all-to-all communication, waiting for the
> caches to load can delay the start of the app, but cache use may decrease the
> overall execution time of the app by more than this delay.  (Loading the cache
> is likely to be more efficient than applications obtaining the path records
> themselves.)
>
> >How multi-pathing is handled in current cache_module?
>
> A kernel ULP can request all paths, then select the one they want.  Beyond that,
> the cache can either return paths to the user round robin or randomly, based on
> the cache settings.
>
> - Sean
>