Date: Tue, 31 Jan 95 13:29:58 MST From: terry@cs.weber.edu (Terry Lambert) To: supervisor@alb.asctmd.com Cc: freebsd-hackers@FreeBSD.org Subject: Re: sys/netipx -Reply Message-ID: <9501312029.AA28146@cs.weber.edu> In-Reply-To: <sf2cedbd.002@alb.asctmd.com> from "supervisor@alb.asctmd.com" at Jan 30, 95 01:44:19 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> Would it make sense to create protocol family callbacks which would be
> available to the ethernet hardware drivers? Is there such an interface
> already defined?
Yes and no.
Callbacks probably mean that you are screwing up load order more than
nything else -- that is, breaking the layering model or otherwise blurring
the distinction between the logical layering.
The "yes" part is that protocols like IPX benefit greatly by peeking into
the packet to know the size before the allocate which cuts the number of
copies by one and (potentially) allows a packet unload directly into the
user buffer.
NetWare uses this internally and has 1/4 the latency in it's stack of the
IPX code in many UNIX implementations ("many" and not "all" because there
are several implementations that have much, much worse latency).
This is very important for request/response architectures, such as SMB
(LANMAN) and NCP (NetWare).
The cost on top of the layering abstraction getting less clean is that
you must have outstanding posted reads to receive a packet, which means
pre-dedicating buffers and other rather uncouth things.
Terry Lambert
terry@cs.weber.edu
---
Any opinions in this posting are my own and not those of my present
or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?9501312029.AA28146>