Path: utzoo!censor!geac!torsqnt!lethe!yunexus!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!cs.utexas.edu!uwm.edu!spool2.mu.edu!uunet!mcsun!unido!fauern!opal!tmpmbx!scuzzy!src From: src@scuzzy.in-berlin.de (Heiko Blume) Newsgroups: comp.unix.sysv386 Subject: Re: binary Mach distribution for 386 Message-ID: <1991Jan15.182524.14868@scuzzy.in-berlin.de> Date: 15 Jan 91 18:25:24 GMT References: <13963@uswat.UUCP> <1991Jan4.140341.11874@granite.cr.bull.com> <1991Jan05.021138.17495@scuzzy.in-berlin.de> <1991Jan5.180851.24 Organization: Contributed Software Lines: 32 shore@mtxinu.COM (Melinda Shore) writes: >In article <1991Jan10.163048.20613@scuzzy.in-berlin.de> src@scuzzy.in-berlin.de (Heiko Blume) writes: >>i think all those intelligent peripherals, like scsi host adapters, multiport >>and ethernet cards, with their own processors and RAM can help a lot with >>these performance problems. >For example, when E&S >was bringing Mach up on their ill-fated supercomputer, they found >that they were only getting 500K throughput on a 10 megabit ethernet. >The problem turned out to be that the folks at CMU had replaced >the software interrupt mechanism for moving packets into the protocol >layer with a kernel thread. The network code, of course, was >pre-tahoe BSD, so they should have been seeing about 1.5 megabits. >You all can do the arithmetic. that's why i mentioned the intelligent peripherals. it's necessary to offload the network code to the interface anyway. for example, with host based tcp/ip over 100Mbit FDDI you get about 20Mbit/s max throughput. if you use the FDDI- and TCP/IP-chips you get about 70Mbit/s. i think that clearly shows the way to go. >And, >if you want to see Mach 3.0 any time in the near future, I would >suggest looking into getting an AT&T source license. just a little bit too expensive... -- Heiko Blume <-+-> src@scuzzy.in-berlin.de <-+-> (+49 30) 691 88 93 public source archive [HST V.42bis]: scuzzy Any ACU,f 38400 6919520 gin:--gin: nuucp sword: nuucp uucp scuzzy!/src/README /your/home