[standards-jig] Pub/Sub for JNG?

Joe Hildebrand jhildebrand at jabber.com
Thu May 2 01:27:32 UTC 2002


Since dave didn't reply directly to any of my valid criticisms of UDP,
I'm taking a slightly less concilliatory tone.  Sorry in advance.

Dave <dave at dave.tj> writes:

> Will somebody please bring up one point that would appear to imply that
> UDP is useless to almost everybody?  If not, then allowing people to
> use UDP if they prefer causes us no harm, and may actually benefit some.

Because (Joe looks around the room) no Jabber implementations will
implement it, unless you write one.  /me holds his breath.

>>  I'll just add one point
>> that hasn't been brought up.  Writing your own protocol on top of UDP would
>> probably be a decent idea if you are a proprietary company with full control
>> over client and server development.  However, for an open source project you
>> will immediately shut off a large number of client developers who don't want
>> to reimplement the protocol in their language of choice....
> The UDP proposal that I came up with (quite off-handedly, as well -
> I'm sure many others can think of even better specs) doesn't reimplement
> most of TCP.  It merely consists of a sequence number for each datagram,
> followed by the actual data.  TCP can't beat that ;-)

Um.  Yes.  It can.  Sliding windows.  Congestion control.

>> especially when
>> they see everything you added to UDP is already in TCP.
> I guess I'd have to admit that I kinda answered that above. . .

I'll have to admit that you didn't.

> ...and it's one that I already answered (and if you don't believe me,
> I'll dig up the URL) - the PPP network that dial-up users connect to
> isn't likely to start requiring congestion control anytime soon, since
> there is only one computer that can transmit on any given channel,
> so there's no contention to contend with :-)

Yet you have to connect to someone on the other side.  If I'm a
service provider, congestion control is a *feature*, since I've
probably got thousands of modems that need to share bandwidth nicely.
No, the PPP link may not need it (which I actually dispute), but the
ISP network *does*.

> Hmm ... that's actually a rather easy quantity to measure: try nslookup
> a few times, and see how many of them work without resorting to TCP.
> (You can just kill TCP at the firewall for a couple of minutes as you
> test, if you want to be sure.)  If you've got a reasonably close DNS
> server, none of them will fail.  The 20% statistic is probably only in
> the congested areas, for long-distance traffic.  With all the Jabber
> servers on the 'net, I'm sure you can find one that'll be able to receive
> most of your datagrams (and if not, just stay with your standard TCP
> connections - I'm not proposing that we eliminate them).  Arguing that
> UDP shouldn't be allowed simply because some people may not be able to
> use it productively isn't very wise, is it?

OK, let me try a different direction.  No, I don't want to force
people to use TCP.  There's just no benefit to server implementors,
administrators, or users, so why would anyone write that code?

> Dave, who wonders how he always rubs people the wrong way. . .

Because you seem to not be willing to allow for the possibility that
other people have thought things through already.

-- 
Joe Hildebrand
Chief Architect
Jabber, Inc.
http://www.jabber.com/




More information about the Standards mailing list