Subject: Re: mbuf external storage sharing
To: YAMAMOTO Takashi <>
From: Jonathan Stone <>
List: tech-net
Date: 10/04/2004 17:42:56
In message <>,
YAMAMOTO Takashi writes:

Yamamoto-san, there seem to be two more-or-less unrelated changes in this patch:

1. Changes to use simple_locks per external mbuf. 
2.  Changes to allocate, and use, a new subflavour of external-storage
mbuf, where the `little' or cluster-header mbuf is embedded in the
external storage.

You seem to be making a bald claim that embedding the `little' mbuf
header in external storage is the `best' approach. Can you provide
evidence to suppor that claim? For it seems to contradict the
conclusions in the FreeBSD-5 paper I cited earlier.

>i looked it.  not very closely though.

Then perhaps you *should* look more closely, if only so we can all
discuss issues of making the network stack SMP-safe from a common,
well-understood, starting point which includes the ``state of the
art'' on fine-grained locking in *BSD network stacks.

[ FreeBSD 5 mbuf rework]

>> I have no idea whether we want to follow the directions in that paper
>> (either mbuf-level API or underlying implemntation); but we should at
>> least think about the quantitative data.
>not relevant to my patch.

Yes and no: it depends very much whether the desired scope of
SMP-safeness is a naive unscalaeable approach, or a more sophisitated
approach that scales to large numbers of CPUs (where "4" has
historically been "large").  And I think that merits a completely
separate discussion.