[standards-jig] Pub/Sub for JNG?

Van Gale cgale1 at cox.net
Wed May 1 23:35:21 UTC 2002


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:
> >
> > --Craig, who likes to optimize at known bottlenecks.

Dave, I think you're missing several important points.  Probably the most
important point you're missing, as evidenced most recently by your signature
quoted above, is that _without measurement_ you are wasting everyone's time.
The point of Craig's signature is _known_ i.e. _measured_ bottlenecks.
Speculation is worthless.  We want to see hard data, not opinion.  If you
can prove total system improvement by using compression then I'm all ears.
If you are just voicing an opinion, especially an opinion that goes against
previous measurements, then you are wasting a helluva lot of people's time.

Now, also regarding the quotes above.  I'm paying the bills for my server,
not the user.  I want the user to have a good experience (otherwise what's
the point of running a server) but I'm not going to increase my real money
costs just so the user can receive a compressed 200 byte message 10 msecs
faster than an uncompressed 300 byte message.  That's what I mean by total
system improvement.

As far as UDP goes, many other people have made critical points that you
seem to be ignoring so I won't go over them again.  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.... especially when
they see everything you added to UDP is already in TCP.

Actually, I will go over a point already mentioned by Joe Hildebrand and
it's one he correctly said was VERY important.

<quote>
- The network will soon begin to require applications to perform congestion
control, and those applications which do not perform congestion control will
be harshly penalized by the network (probably in the form of preferentially
dropping their packets during times of congestion).
</quote>

UDP datagrams are the first to go everytime.  20% of UDP packets dropped at
the metropolitan access points (MAE, etc.) is a good day.  This means the
reliability you build on top of UDP is going to have to work much much
harder than TCP.

Van





More information about the Standards mailing list