hb@lln-cs.UUCP (Hubert Broze) (08/01/88)
In the past (augustus 87), I read in the NEWS (comp.unix.questions), that some people have encountered some problems on a MicrovaxII with 2 DEQNA's on Mt-Xinu 4.3. I have just configured my MicrovaxII as a gateway machine forfor TCP/IP with 2 DEQNA's and I have the following problem: when the gateway is active, it runs fine but all other activites (on local disks) are impossible. It would like to meet (by mail) people who have some experiences with the problem mentionned above. Thanks in advance, H. Broze System Manager E-mail: prlb2!lln-cs!hb (Belgium)
lee@turing.unm.edu (Lee Ward) (08/11/88)
In article <870@lln-cs.UUCP> hb@lln-cs.UUCP (Hubert Broze) writes: >In the past (augustus 87), I read in the NEWS (comp.unix.questions), >that some people have encountered some problems on a MicrovaxII with 2 DEQNA's on >Mt-Xinu 4.3. I have just configured my MicrovaxII as a gateway machine forfor TCP/IP >with 2 DEQNA's and I have the following problem: when the gateway is >active, it runs fine but all other activites (on local disks) are impossible. > Wow, an exact description of the problem we have been having for about a year. Your problem is more extensive than you know. 1) The splnet() in your kernel could be wrong. Though if you have gotten to the "my disk quits responding" part it is most likely right. Just in case here is a diff of /sys/vax/inline/langpats.c RCS file: RCS/langpats.c,v retrieving revision 1.1 diff -r1.1 langpats.c 41c41 < mtpr $0x16,$18\n" }, --- > mtpr $0x17,$18\n" }, 2) You can REALLY aggravate the problem if you have two disk controllers present. When IO activity gets heavy the system freezes. If you have a systat (:v) running you will see your load go up, up, up and nobody but the systat will come on the run queue. A kludge is to put a timeout in your driver that will kick the controllers. If you do this and try to run both DEQNA's your disk controllers will get kicked alot but, alas, it still won't work. I have a program that I can send to you that will freeze your system every time. I won't post it as some malicious user might want to try it too. If ANYONE can get two DEQNA's going under MT-XINU please let me know too. We're about to put up Ultr*x to solve the problem. We have had to move 20 machines over to the campus net and they page like mad. Our floor loop is entirely useless until we can bring the second DEQNA on line. --Lee (Ward)
ellis@godot.psc.edu (James Ellis) (08/13/88)
In article <1169@unmvax.unm.edu> lee@turing.unm.edu (Lee Ward) writes: >In article <870@lln-cs.UUCP> hb@lln-cs.UUCP (Hubert Broze) writes: >>In the past (augustus 87), I read in the NEWS (comp.unix.questions), >>that some people have encountered some problems on a MicrovaxII with 2 DEQNA's on >>Mt-Xinu 4.3. I have just configured my MicrovaxII as a gateway machine forfor TCP/IP >>with 2 DEQNA's and I have the following problem: when the gateway is >>active, it runs fine but all other activites (on local disks) are impossible. >> > When this happened to us the problem turned out to be that the ethernet driver was reserving too many receive buffers. Mike Karels suggested lowering NRCV. (NRCV is in vaxif/if_qe.c.) We dropped it from 25 to 15 on our systems and have had several (busy) such gateways working fine since. This has worked on both UCB 4.3 and Mt. Xinu. Jim Ellis Pittsburgh Supercomputing Center