<< | Thread Index | >> ]    [ << | Date Index | >> ]

Subject: Re: Proposal: Compression of large packets
From: Josef Drexler <jdrexler,AT,josefsbox,DOT,cjb,DOT,net>
Date: Mon, 3 Dec 2001 19:42:13 +0100
In-reply-to: <Pine.LNX.4.21.0111301455020.22203-100000@raid.kaico.com>

On Fri, 30 Nov 2001 ewheeler,AT,kaico,DOT,com wrote:

> I agree: We should find out which bit we could use for zpackets from
> Olaf.  For the moment, would it be kosher to just use one arbitrarily for
> dev purposes?

It'll have to be...

> For configuring the compression of packets, there should be an option that
> reads: if length(zpacket) is X% less than length(packet), send
> zpacket; else send packet.

Actually, there is no sense in not sending a compressed packet if the
length is even 1 byte less than the uncompressed length.  Decompression is
very fast.  My suggestion was however that we should not attempt to
compress small packets, since the potential benefit is small compared to
large packets.  Only very few of the small packets can be compressed

> This would add considerable latency, but it could be coupled with 'and
> length(packet) > 100' or some such thing.  It may also make sense to check
> the TOS flag.  If the packet is flagged 'Minimize-Delay 16 (0x10) (from
> iptables)', then never compress.  That way we do our best to compress
> everything, and let minimum delay packets through as fast as possible.

That's a good idea, I wonder how that would work in conjunction with
ipchains/iptables.  Would it be possible to mark a packet in one of the
tables or chains before CIPE gets to tunnel it?  That way you could
disable compression of packets based on port numbers, like ssh packets for

By the way, I think it would be useful to move the discussion of
implementation details to private emails.

   Josef Drexler                 |    http://publish.uwo.ca/~jdrexler/
 Please help Conserve Gravity    | Email address is *valid*.
 Go by car/train, don't fly.     | Don't remove the "nospam" part.

<< | Thread Index | >> ]    [ << | Date Index | >> ]