Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 26 Sep 2003 12:18:17 +0200
From:      Marko Zec <zec@tel.fer.hr>
To:        Roderick van Domburg <r.s.a.vandomburg@student.utwente.nl>
Cc:        freebsd-net@freebsd.org
Subject:   Re: DEVICE_POLLING together with link0 interrupt offloading?
Message-ID:  <3F741269.3A2FF9FC@tel.fer.hr>
References:  <000501c3840c$6da6ae60$6ba55982@gog>

next in thread | previous in thread | raw e-mail | index | archive | help
Roderick van Domburg wrote:

> Hello everyone,
>
> Just curious: I just noticed that DEVICE_POLLING works perfectly well with
> the interrupt offloading feature of the fxp driver (link0). In my
> understanding, both try to improve ``network scheduling'' for lack of better
> words - especially under load. Also I now understand that DEVICE_POLLING is
> preferred over interrupt offloading, especially on routers.
>
> But just for the sake of curiosity: what does enabling _both_ bring about to
> the network performancy and latency? I'm thinking that latency would be
> greatly increased because the poller would often miss the bundled packets.
>
> What's your take on this?

There's no much sense in enabling both the network interface polling mode and
fxp interrupt coalescing microcode at the same time, since the polling code
disables all network interface interrupts anyway. Moreover, there wouldn't be
any additional latency, since the polling routine would extract all incoming
packets from inbound DMA rings as soon as they would arrive.

Marko




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3F741269.3A2FF9FC>