lawley@muluru.cs.mu.OZ.AU (michael lawley) (08/17/90)
We are running a Multimax 320 with 64Meg of physical memory, 2 EMC's, 18 CPU's (APC's) and approx 600Meg of swap spread evenly across 5 disks. Recently, our student load has increased to about 80 students; a total of about 100 users at a time. (The machine is also doing some NFS serving and running Macintosh CAP software for about 50 Macs.) The problem we are experiencing is very poor response from the system. This seems to happen very suddenly, as if the system hits a brick wall. But the load on the machine is only around 5 i.e. we have 13 idle CPU's. These appear to be the symptoms of thrashing. Are there any (sysparam) parameters we should look at or system statistics we can examine to try and track them problem down. Any help would be much appreciated. mike -- _--_|\ michael lawley (lawley@cs.mu.OZ.AU). / \ The Unicycling Systems Programmer, \_.--.*/ Melbourne University, Computer Science v
ted@gouldnl.encore.nl (Ted Lindgreen) (08/17/90)
In <lawley.650871623@muluru> lawley@muluru.cs.mu.OZ.AU (michael lawley) writes: > ... Are there any (sysparam) >parameters we should look... You could take a look at Dynsize: increase it by a factor of 1.5 for instance. Also look at Maxusers: never make this parameter too high, to accomodate some 80 students, I expect that 40 to 50 is enough. And check Max_procs (lower it to something reasonable). At a site in Holland there were similar problems, which disappeared after changing: Maxusers (from 100 to 50) Max_procs (from 50 to 30) Dynsize (from 524288 to 750000) At that time (they have upgrade their system with XPC's later), their system was about half the size of your system, but their usage (also mostly students) was more than half of yours. -- | Ted Lindgreen ted@encore.nl | | Encore Unix Centre Europe ...!mcsun!hp4nl!encore.nl!ted | | Maarssenbroek, The Netherlands (USA) tlindgreen@encore.com |
arthur@otto.bf.rmit.oz.au (Arthur Adamopoulos) (08/21/90)
In article <lawley.650871623@muluru>, lawley@muluru.cs.mu.OZ.AU (michael lawley) writes: > We are running a Multimax 320 with 64Meg of physical memory, > 2 EMC's, 18 CPU's (APC's) and approx 600Meg of swap spread > evenly across 5 disks. > The problem we are experiencing is very poor response from the This sounds like two different problems we have had in the past (UMAX V). When we had all our disks on 2 EMC cards we were getting _very_ bad slowdown when we got about 30 users logging on. This problem disappeared when we got our MSC. We now have one MSC handling all disks & tape and one EMC handling the ethernet. I would suggest doing a 'Csar -bd' while you are getting the slow performance and have a look at the load levels etc. on the disks. CPU isn't the only factor that could cause poor response. If the harddisks are overloading the EMCs then the comms will also slow down. We have also in the past had a problem of the entire machine not allowing any new processes to start but still processing any existing processes. The reason for this was never found but it hasn't happened for a number of months. This seemed like a 'brick wall' effect. ----------------------------------------------------------------------- Arthur Adamopoulos ACSnet: arthur@otto.bf.rmit.oz Business Info Systems Internet: arthur@otto.bf.rmit.oz.au RMIT 411 Swanston st. Tel: +61 03 660 3470 Melbourne 3000 ICBM: 37 48 S / 147 58 E -----------------------------------------------------------------------