Re: Dom0 detects no carrier on ixg

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Wed, Dec 14, 2022 at 05:44:15PM -0700, Brook Milligan wrote:
> 
> > On Dec 14, 2022, at 7:54 AM, Manuel Bouyer <bouyer%antioche.eu.org@localhost> wrote:
> > 
> > On Wed, Dec 14, 2022 at 07:40:53AM -0700, Brook Milligan wrote:
> >> 
> >>> On Dec 14, 2022, at 1:39 AM, Manuel Bouyer <bouyer%antioche.eu.org@localhost> wrote:
> >>> 
> >>> 55 is ENOBUFS; you don't have enough memory, or more probably enough
> >>> mbuf clusters
> >>> 
> >>> NMBCLUSTERS is computed automatically based on RAM amount, but for a
> >>> dom0 (which typically has a small portion of the physical RAM allocated),
> >>> it ends up too low for modern NIC.
> >>> 
> >>> I use
> >>> options NMBCLUSTERS=262144
> 
> OK, I?ve increased this to 4M (4,166,656) and RAM to 8GB; no more ENOBUFS errors.
> 
> These seem like quite high values.  Do I just fiddle to see what is required?  Is there any rule of thumb for figuring this out?

Most of it is for NIC receive queues. I have in mind 64k per ixg(4) instance
(but I may not remember properly) + 64k/128k for general use (depending
on usage)

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--