[ofa-general] Re: [PATCH 2 of 3] IB/mthca: support multiple completion vectors

[Michael S. Tsirkin]

> Quoting Roland Dreier <rdreier at cisco.com>:
> Subject: Re: [PATCH 2 of 3] IB/mthca: support multiple completion vectors
> 
>  > > I think #vectors == O(#CPUs) is what we should aim for.
>  > 
>  > What do you suggest to do about memory scalaibility?
> 
> Right now we default to 64K CQs, so each completion EQ uses 64K * 32 bytes,
> which is 2 MB.  Which is a lot but not a killer on modern servers.
> And I would expect memory to scale as O(#CPUs) too.  So I think 1 or 2
> completion EQs per CPU is the right amount.

With dual-core, I'm not sure memory scales as fast as #CPUs anymore.

> I guess it would be nice if there was something tricky we could do to
> adjust things on the fly but I don't think it's worth getting too tricky.

Problem is, #CPUs is not a static value anymore.
With CPU hotplug num_possible_cpus is quite often 4 or 8 even though
one actually has a single one present.

-- 
MST