Date: Tue, 13 Oct 1998 09:43:48 +0100 From: Brian Somers <brian@Awfulhak.org> To: Bruce Evans <bde@zeta.org.au> Cc: brian@Awfulhak.org, kjc@csl.sony.co.jp, current@FreeBSD.ORG Subject: Re: Improper sharing of modem bandwidth Message-ID: <199810130843.JAA00674@woof.lan.awfulhak.org> In-Reply-To: Your message of "Tue, 13 Oct 1998 15:41:01 %2B1000." <199810130541.PAA12787@godzilla.zeta.org.au>
next in thread | previous in thread | raw e-mail | index | archive | help
> >Surely, once the packet queueing delay is reduced, we need to reduce
> >the tty delay. The current kernel doesn't have ioctls for the
> >buffer size but slip uses a smaller tty buffer and has a better
> >response time than ppp.
>
> User ppp is behind a _lot_ of output buffering, including (at speeds >=
> 57600 Kbps) a 4KB tty buffer (up from 2KB in FreeBSD-2.2). Perhaps the
> increase from 2KB to 4KB is especially significant when the MTU is 1500.
Heh, I *assumed* (yes, I know) that the tty buffer was around 128 or
256 bytes, but definitely less than 1500 !!!
So in practice, in 3.0 with a 1500 MTU, we won't be able to skip ahead
of the 4096 bytes (assuming a busy link) at the front of the queue.
Without suspend/resume, we'll have to wait behind up to 1500 additional
bytes (750 avg) and with suspend/resume we'd wait behind an
additional 0 bytes and send 1 or 2 additional HDLC escape codes to do
the suspend/resume. This gives a ~15% latency improvement, which
isn't brilliant, but is better than nothing.
If there was an ioctl to change the usable tty buffer size to less than
4k, ppp could set it to ``whatever the MTU is'' and get quite a
reasonable interactive performance gain.
> Kernel ppp doesn't have this problem. Kernel ppp also has a fast path
> for incoming packets to avoid the average 5ms latency for polling the
> pseudo-DMA input buffer. This only seems to be important for local
> connections.
>
> Bruce
--
Brian <brian@Awfulhak.org>, <brian@FreeBSD.org>, <brian@OpenBSD.org>
<http://www.Awfulhak.org>;
Don't _EVER_ lose your sense of humour....
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199810130843.JAA00674>