Date: Mon, 06 Nov 2000 08:18:40 +1100 From: Peter Jeremy <peter.jeremy@alcatel.com.au> To: Sam Zamarripa <samz@oz.net> Cc: Gregory Bond <gnb@itga.com.au>, freebsd-stable@FreeBSD.ORG Subject: Re: PPP Nat Bandwidth Sharing Message-ID: <00Nov6.091808est.115209@border.alcanet.com.au>
next in thread | raw e-mail | index | archive | help
On Mon, Oct 30, 2000 at 12:26:16PM +1100, Gregory Bond wrote:
>> Bandwidth sharing doesn't seem to be too efficient. For example if Machine-1
>> on the LAN is doing a major download of a compressed file and getting near
>> full speed of 5.6K/sec, if Machine-2 starts web browsing or a download
>> itself..they can barely pull 1K/sec.
...
>Secondly, with a typical loaded 56k setup, there are several seconds of data
>queued up already so bandwidth sharing is not likely to be effective for a new
>request (i.e. browsing from Machine-2) until all the queued data has drained -
>which is too late to be useful.
Actually, you do have some control over this. net.inet.tcp.recvspace
controls the TCP receive window size advertised by your machine.
Reducing this will reduce the amount of TCP data queued between the
TCP connection end-points. There are several caveats:
1) It has no effect on UDP, ICMP etc
2) It limits the queue size for each TCP connection rather than your
total queue size at the ISP. This reduces its effectiveness if there
are lots of TCP connections, but if you are just trying to browse in
one window whilst waiting for an FTP in another, it's quite effective.
3) It has to be set in the end-point machine, not the firewall/NAT box[1]
4) It affects all network connections to the machine
5) TCP connection bandwidth is limited to (roughly) windowsize/RTT, so
the value is a tradeoff between responsiveness and bandwidth to
distant sites.
6) Points 3-5 mean LAN performance may be reduced slightly.
[1] Actually, if you wanted to be really creative, you could probably
doctor the NAT code to also reduce the advertised window size. The
code to do this is left as an exercise for the reader.
Peter
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?00Nov6.091808est.115209>