pop@linus (01/12/90)
I was wondering if people thought that using X over a long-haul
network (T1) was an adequate design for an operational system or
whether it would be just infeasible.
In reading some of the early X doc (Scheifler '86) it was stated that X
was intended for networks with a 50ms roundtrip time, or a campus wide
net. But from Boston I seem to ping Washington D.C. in 40ms. and
Colorado in 60ms (T1 link). I don't know if that round trip time is
sustained, but in running some sample X clients at such distances did
not seem to be a problem. The limitation was probably the server on
my end (Sun3 X11R3).
It would seem to me that applications that have little user input
(cut down on those round trip times) would work well under X. The main
advantage is that I would not have to come up with an application
specific protocol to support the distribution of graphic information
to remote sites. Any comments or experiences with doing this ?
Please email direct, I have a hard time keeping up with the volume.
Thanks, Paul.
Paul O. Perry MITRE Corporation
Phone: (617) 271-5641 Burlington Road
ARPA: pop@mbunix.mitre.org Bedford, MA 01730
UUCP: ...{decvax,utzoo,philabs,security,allegra,genrad}!linus!popjordan@morgan.COM (Jordan Hayes) (01/17/90)
# why do i send mail to this list when all i get is about 40 bounces? Paul O. Perry <pop@mbunix.mitre.org> asks: I was wondering if people thought that using X over a long-haul network (T1) was an adequate design for an operational system or whether it would be just infeasible. I regularly use X over a 9.6k asynchronous line from home, and I don't think it's "infeasible" -- I often have several clients running that probably saturate the network (SLIP between a Sun and Cisco using V.32 on a Telebit), but it feels okay to me. I also remember playing xtrek over T1 with good results (altho I got blown away a lot because of the latency -- they shot me well before I saw them ;-( /jordan
nmm@cl.cam.ac.uk (Nick Maclaren) (01/17/90)
I did not see the original question, but a follow-on had this: > Paul O. Perry <pop@mbunix.mitre.org> asks: > I was wondering if people thought that using X over a long-haul > network (T1) was an adequate design for an operational system > or whether it would be just infeasible. Some measurements by people at Edinburgh and elsewhere indicate that you need to transfer about 9.6Kbits/second in both directions to achieve reasonable response. This is a problem for those of us where dial-up means 2400/2400 at most (and often 1200/75), but is not a design problem with X. Ditto requirements for local store, fast access to a font server, etc. The fundamental design problem that I know of (in X11.3, at least) is this: The X design requires at least one packet to do the round trip from the keyboard to the application and back per character typed, and may require four (or, rarely, more). If the application supports auto-repeat, the minimum is two. This must happen BEFORE reflection. The people at Edinburgh measured about 5 packets/second for ordinary users doing ordinary work. Fast typists or graphics work could easily be 5-10 times as much, depending on the use and application. This is a serious strain on almost all switching systems, and is likely to remain so for the forseeable future. If you are sufficiently unfortunate as to want to use your local X-server to a client in a far country (e.g. from Europe to the U.S.A.) and get switched via a satellite, there is some 0.6 second delay on each character reflected (5 characters at a normal fast typing speed). With the X11.3 design, this is fundamental and cannot be removed without changing the speed of light! Nick Maclaren University of Cambridge Computer Laboratory nmm@cl.cam.ac.uk