twolf@homxb.ATT.COM (T.WOLF) (03/01/89)
A lot of questions have been posted regarding X performance over low-speed dial-up lines. I have a more 'mundane' question :-) Has anyone collected any data on the number of X-Servers which can be supported (reasonably) on a LAN running at 1, 5, 10, or 100Mbps? More specifically, if the LAN consists of a single "client-server" host and multiple X-servers, how many concurrent X-clients does it take to slow down I/O response time beyond tolerable limits? Also, at what speeds does the single host have to operate in order for the network to remain the bottleneck? I know that words such as "reasonable" and "tolerable" don't have a whole lot of meaning -- I leave it up to whoever has answers to decide that :-) I guess what I'm really asking is whether anyone has performed any performance analyses in these areas. If you can point me towards papers, theses, articles that have been published on the subject, I'd appreciate it. -- Tom Wolf Bell Labs, Holmdel, NJ E-mail: twolf@homxb.att.com (My employer doesn't know about these and other incriminating remarks)
ed@lupine.UUCP (Ed Basart) (03/03/89)
In article <3079@homxb.ATT.COM>, twolf@homxb.ATT.COM (T.WOLF) asks
if anyone has collected data on the number of X-Stations can be reasonably
supported on a LAN, especially just considering one host machine and a bunch
of X stations.
I had replied to this previously, but I'll do it again. Our experience at
NCD was to attach 10 NCD X stations to a Sun 4/260, then put a worst
case load on the Sun and on the network. The simplest case turns out to be
running ICO or else "cat"-ing a file to the screen. With 10 stations the
Sun was at 100% cpu utilization with about 650 packets/sec observed on the
network.
The limitation in this case appears to be the Sun's cpu power, not the
network. Other tests seemed to indicate that the Sun is limited in the
number of packets per second that it can handle. Whether the actual
bottleneck is in the network software, the network hardware, or the
rumored "problems with context switching" in the Sun 4 is unknown at
this point.
--
Ed Basart
Network Computing Devices
uunet!lupine!edtwolf@homxb.ATT.COM (T.WOLF) (03/04/89)
In article <213@lupine.UUCP>, ed@lupine.UUCP (Ed Basart) writes: > > I had replied to this previously, but I'll do it again. Our experience at > NCD was to attach 10 NCD X stations to a Sun 4/260, then put a worst > case load on the Sun and on the network. The simplest case turns out to be > running ICO or else "cat"-ing a file to the screen. With 10 stations the > Sun was at 100% cpu utilization with about 650 packets/sec observed on the > network. > Thanks for the reply. I didn't receive your previous one. In the case you mentioned, the Sun was the bottleneck. I guess I'd also like to know more about how many X-terminals could be "serviced" iff the host machine wasn't the restricting element. From what I understand, CSMA/CD-based lans like Ethernet quickly bog down when traffic becomes heavy (collisions and such.) Maybe this question is best asked in another newsgroup (does a "comp.lan" exist?) But experts on LANs may not know the "traffic-patterns" of X-clients. -- Tom Wolf Bell Labs, Holmdel, NJ E-mail: twolf@homxb.att.com (My employer doesn't know about these and other incriminating remarks)