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

Subject: Re: Packet compression
From: ewheeler,AT,kaico,DOT,com
Date: Wed, 30 May 2001 06:09:55 +0200
In-reply-to: <004401c0e7b7$08614e00$0344a8c0@flexsysgroup.com>

On Mon, 28 May 2001, Allan Latham wrote:

Allan -- 
        Could I get the cipe compression code?  Could you send me a url to
it? There's no sense in reinventing the wheel if you've done it.  
What is the draw back to a module that takes a lot of kernel memory?  Does
it slow the system down?  At this point memory is cheap (~$30 for a 128mb
stick!)  I realize that tight code reduces bloatware but if it means an
increase in overall performance I'm always for that!

 The other question is about VJ style header compresion.  How would
you remove redundant ip/tcp/udp headers?  
My average ping time to the host over the internet is ~57.9ms and my avg
ping over the CIPE VPN is 63.9ms.  How much would/could cutting redundant
headers help?  What else can help cut the latency?

Latency is our biggest problem.  I'm running a sql based problem over VPN
lines using CIPE.  Since the ODBC protocol being used is very poorly
written it waits for confirmation for each query request (this is really
silly considering it uses a TCP port for communication and TCP is by
definition guaranteed communication).

Any ideas to help speed stuff up would be awesome.

--Eric  

> I does help (on average data) - about 50% can be normal for web pages and
> plain text etc.
> 
> BUT:
> 
> 1. It's heavy on kernel memory
> 2. It's ineffective on much binary data
> 3. It's something else to go wrong
> (4. It uses CPU cycles - not a problem any more!)
> 
> I implemented this on a much earlier version of cipe - it's way out of
> date - but we have used it for a couple of years trouble free.
> 
> Much more interesting is packet compression by removing redundant ip/tcp/udp
> header fields (a bit like VJ). This does not increase throughput much but it
> has a massive effect on latency and hence responsiveness of interactive
> sessions.
> 
> Best regards to all
> 
> Allan
> 
> 
> 
> 
> ----- Original Message -----
> From: <ewheeler,AT,kaico,DOT,com>
> To: "Karl Kleinpaste" <karl,AT,charcoal,DOT,com>
> Cc: <cipe-l,AT,inka,DOT,de>
> Sent: Monday, May 28, 2001 1:21 AM
> Subject: Re: Packet compression
> 
> 
> > On 27 May 2001, Karl Kleinpaste wrote:
> >
> > > ewheeler,AT,kaico,DOT,com writes:
> > > > Would there be any appreciable performance increase over low bandwidth
> > > > lines if the packets were compressed using zlib before they were
> > > > encrypted?
> > >
> > > It doesn't do that much good.  But it helps a little.
> > >
> > > As a comparison point, experiment a bit with ssh, using different
> > > compression levels in .ssh/config, e.g.
> > >
> > > Host your.chosen.test.destination.host
> > >   Compression yes
> > >   CompressionLevel 9
> > >
> > > Try various digits, and do some large file copying over your slow link
> > > with scp.  Time the results.
> > >
> >
> > I know this can help alot w/ ssh on plain text and the like but I was
> > wondering specifically about CIPE.  Could this be implemented inthe packet
> > encapsulation and would we see a performance increase accross low
> > bandwidth lines.    My guess is that it would;  I just took a block of
> > text:
> >
> > [root@dev dev]# dd if=/var/log/messages bs=1400 count=1 | gzip -v9 | wc -c
> > 1+0 records in
> > 1+0 records out
> >  77.7%
> >     330
> >
> > You can see that even for mall 1400 byte packets we can get a 77% size
> > cut.  Would it be feasible to implement this into CIPE?
> >
> > --Eric
> >
> >
> > --
> > Message sent by the cipe-l,AT,inka,DOT,de mailing list.
> > Unsubscribe: mail majordomo,AT,inka,DOT,de, "unsubscribe cipe-l" in body
> > Other commands available with "help" in body to the same address.
> > CIPE info and list archive:
> <URL:http://sites.inka.de/~bigred/devel/cipe.html>
> >
> 
> 
> --
> Message sent by the cipe-l,AT,inka,DOT,de mailing list.
> Unsubscribe: mail majordomo,AT,inka,DOT,de, "unsubscribe cipe-l" in body
> Other commands available with "help" in body to the same address.
> CIPE info and list archive: 
><URL:http://sites.inka.de/~bigred/devel/cipe.html>
> 





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