[ewg] Re: [PATCH] IB/ipoib: increase ring sizes

[Eli Cohen]

On Wed, 2008-06-04 at 21:13 -0700, Roland Dreier wrote:
> > Increase IPoIB ring sizes to twice the original size to act as
>  > a shock observer for high traffic picks.
First let me correct myself, absorber and not observer.

> 
> Looks fine but I would like to include a little more motivation in the
> changelog.  What type of workload benefits from this change?

So this is what I suggest for the changelog:

Increase IPoIB ring sizes to twice the original size to act as
a shock absorber for high traffic picks. With the current settings, we
have seen cases that there are many calls to
netif_stop_queue()/netif_wake_queue(). This causes degradation in
throughput. Also larger receive buffer sizes help IPoIB in CM mode to
avoid experiencing RNR NAK conditions due to insufficient receive
buffers at the SRQ. This increase is still modest compared to other
Ethernet drivers -- bnx2 uses 256 for it HW queue length (on 4K page
size machines).