[standards-jig] Stream Compression
julian at jabber.org
Thu May 2 15:12:18 CDT 2002
Dave, you need to stop stating things as facts when you obviously have
not done any research on the topic whatsoever. I just typed "SSL
compression" into google and I got a bunch of documents on how you
enable compression in SSL.
In fact, there's a nice little page on the current situation of
compression in SSL using the OpenSSL library:
TSLv1, the method Gabber uses, supports compression. It's simply a
matter of getting client and server to negotiate the same compression
method. After agreeing upon a standard set of compression IDs for
Jabber, it would take probably 1-3 lines of code in Gabber and maybe a
few more in the server to get compression enabled in Jabber SSL
sessions. What Mike Lin describes would take a heck of a lot more work
Please, Dave, while I appreciate the amount of time you spend on Jabber,
the message you just sent was utterly worthless because you really don't
know the facts. You were just telling me I was wrong for the sake of
telling me I'm wrong.
Also, before you argue that SSL adds the encryption overhead, jer just
pointed out to me that it's probably possible to tell SSL to use a null
cipher, so you essentially get what Mike Lin described. Looking at the
OpenSSL headers and documentation, it seems possible, but I will admit
that I haven't tried it.
On Thu, 2002-05-02 at 15:34, Dave wrote:
> This isn't about UDP, so I believe I'm allowed to reply to your message:
> SSL doesn't compress; it only encrypts.
> - Dave
> Julian Missig wrote:
> > http://mailman.jabber.org/pipermail/jdev/2002-January/009560.html
> > Is Mike Lin's post on gzipping Jabber. He put together code tests and
> > everything... and his conclusion was that once you put together all the
> > pieces needed to make gzip into something decent for Jabber, you're very
> > close to SSL (so why not just use SSL?). See other messages in the
> > thread for further information.
> > Julian
> > On Wed, 2002-05-01 at 18:23, Dave wrote:
> > > Dial-up users also like to download data, you know. . .
> > >
> > > - Dave ... who likes to let people optimize at their own bottlenecks :-)
> > >
> > >
> > > Craig wrote:
> > > >
> > > > <gzip-over-udp-madness snipped>
> > > >
> > > > I know this is obvious, but I just _have_ to chime in with it. I've
> > > > done a fair amount of load testing of JCS in a lot of different
> > > > configurations. Load tests are typically done on a 100Mbit network
> > > > (shared, not segmented) -- probably less than you'd have on the
> > > > backplane of a serious install. Running at seriously high throughput,
> > > > it turns out that processor -- not network is the bottleneck for a mix
> > > > of JUD/TC/IM traffic. Couple this with the exceptional job that modern
> > > > network hardware does at text compression and I'm left wondering what
> > > > the advantages in udp-as-tcp-with-gzip are here? Are folks running
> > > > servers with thousands of users at home on the other side of a dialup?
> > > > Are end users having a problem with the couple hundred bytes going
> > > > through every so often? Are people really having problems with the
> > > > bandwidth? Am I just not getting it?
> > > >
> > > > --Craig, who likes to optimize at known bottlenecks.
More information about the Standards-JIG