This bandwidth issue will be with us forever.  How about adding a short
section of "Advice to implementors and users" to the render protocol
description.  Or it could be a separate paper with a reference in the
protocol.  But something so that a common initial understanding can be
built up.

There are two related performance issues:
 1) bits transmitted - it is useful to minimize these, 
 2) synchronous round trips - these are orders of magnitude more
 costly than bits transmitted.  It is worth sending 10-100x more bits
 if there is a corresponding reduction of round trips.

An explanation of how to send glyphsets efficiently, reference them
efficiently, and indicative numbers on how display of a page of text
translates into bits and round trips would then orient people.  It would
also help eliminate poor implementations by new implementators and
applications.

I think that including realistic numbers would help a lot.

As for networks, there is a real need to keep X usable over network
links.   I personally find something like 90% on machine / 9% over LAN /
1% over WAN.  I would hate to lose that 9% or 1% because even though
they are a small part of my usage, they are an important part of my job.

WAN link speeds are not going to improve dramatically over the next
decade or two.  The speed of light problems are intractable, and the
installed infrastructure changes slowly.  True, I can upgrade a clean
fiber to 140Tb/sec.  But the real problem is how fast can I transmit
from office A at site A to office B at site B, for thousands of sites.
That speed limit changes slowly.  It may even decrease on average with
people exploiting wireless and cable technologies that trade bandwidth
for flexibility, cost, and distance.  Nothing will change the 10-20ms
minimum for a round trip. That is a speed of light limit, and with most
routing paths you are hard pressed to even get close to that limit.

Typical LAN speeds will probably deteriorate.  Again, this is an
infrastructure issue.  With an expensive infrastructure I can put in
1 Gb/s LAN.  For real cheap I can use wireless, phone cable, or 100
Mb/s.  Those other technologies are mostly down in the 1-10 Mb/s range.
Cost and flexibility will drive increased use of the slower
technologies.

I think that we should target: 
 - usable, fast text using non-AA fonts over 50/25 kbps with o(10)ms
   round trip.  (I'm using client/server transmit speed notation.  50/25
   could be a 56K modem or a low power wireless data.)
 - borderline, sluggish text using AA fonts over 50/25.
 - usable, fast text using AA fonts over 768/128 (e.g. DSL, cable, wireless)
 - immediate text, OK graphics and images over 10,000/10,000
 - immediate text, fast graphics and images over 100,000/100,000.

>From what I've seen the current X and Render do meet these goals with
at least some applications.  So I think that my request is much more one
of explaining what has already been accomplished and then avoiding new
problems as it evolves.

R Horn


_______________________________________________
Render mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/render

Reply via email to