twolf@homxb.ATT.COM (T.WOLF) (01/10/89)
This has probably been discussed previously ... but since I just started reading this group... :-) What is the average length (if there is such a thing) for an XProtocol packet on a TCP/IP (Ethernet)? Does Xlib stuff multiple requests into a single packet which it sends to the server? I know that Xlib buffers up these requests (but does it send each as a packet?) Sorry iff these questions are naive. On a similar note: What is the lowest-bandwidth network X has been ported to? Does someone know whether it is feasible (performance wise) to run X over a 9.6kbit connection? Well, I've exhausted my nearly infinite store of questions -- for now. Thanks in advance for any answers. Tom Wolf Bell Labs, Holmdel, NJ E-mail: twolf@homxb.att.com
jg@jumbo.dec.com (Jim Gettys) (01/11/89)
In article <2772@homxb.ATT.COM> twolf@homxb.ATT.COM (T.WOLF) writes: > >This has probably been discussed previously ... but since I just started >reading this group... :-) > >What is the average length (if there is such a thing) for an XProtocol packet >on a TCP/IP (Ethernet)? Does Xlib stuff multiple requests into a single >packet which it sends to the server? I know that Xlib buffers up these >requests (but does it send each as a packet?) Sorry iff these questions >are naive. Xlib buffers requests before sending, so for most applications full size packets are sent on the network. Multiple requests are sent in a single write (packet, or whatever the transport cares to do). > >On a similar note: What is the lowest-bandwidth network X has been ported to? >Does someone know whether it is feasible (performance wise) to run X over a >9.6kbit connection? People have played with 9.6 lines; performance is marginal. With data compression/caching schemes, things improve greatly. Hopefully this work will see the light of day someday. We've seen 2X to 5X improvement in some tests. At 19.2, lots of things are reasonable even without such techniques; most things run pretty well at 56kbaud. Of course, some of us are used to more than 50,000-70,000 characters per second (DECstation 3100, announced today), and the idea of 1k/second is a bit slow. :-) - Jim Gettys Digital Equipment Corporation Cambridge Research Laboratory Cambridge Mass.