saus@media-lab.media.mit.edu (Mark Sausville) (10/08/89)
Hardware: VAX 6320 OS: Ultrix 3.1 How do I find out how my system is doing in regard to the utilization of physical memory in the machine? vmstat has the following output: procs memory page disk faults cpu r b w avm fre re at pi po fr de sr r1 r2 r3 r4 in sy cs us sy id 1 0 0 36k 184k 0 1 1 0 0 0 0 0 0 0 0 20 100 18 11 4 85 The memory fields seem to be fairly consistent around the values shown even when the load is high. This machine has 128 MB of main memory and has ~160 MB ow swap configured. Pstat -s shows: 158361k swap configured 33471k used (2175k text, 0k smem) 124888k free, 6999k wasted, 2k missing My guess would have been that the maximum amount of memory which can get allocated is 128M (phys. mem) + 160M (swap) less kernel space. Is this right? I have never seen the avm field anywhere close to 128k. Why does the system swap/page when there seems to be lots of free physical memory? I suspect I'm showing a good deal of ignorance here, but it's true, I am ignorant. I'm new to both VAX architecture and Unix/Ultrix guts. I'm willing to rtfm, but would prefer a brief elucidation, with references. The man pages for vmstat and pstat are more than a little cryptic to me. And another thing: I suspect I'm not getting much out of the second processor. Has this been thrown around before? A while back, when troubleshooting we turned off the 2nd cpu and didn't notice much of a hit. Any pointers to understanding how the 2nd proc gets allocated tasks and how effective it is? Ignorance is NOT bliss, Mark. Mark Sausville MIT Media Laboratory Computer Systems Administrator Room E15-354 617-253-0325 20 Ames Street saus@media-lab.media.mit.edu Cambridge, MA 02139
alan@shodha.dec.com ( Alan's Home for Wayward Notes File.) (10/09/89)
In article <SAUS.89Oct8103308@media-lab.media.mit.edu>, saus@media-lab.media.mit.edu (Mark Sausville) writes: > > Hardware: VAX 6320 > OS: Ultrix 3.1 > > How do I find out how my system is doing in regard to the utilization > of physical memory in the machine? vmstat has the following output: > > procs memory page disk faults cpu > r b w avm fre re at pi po fr de sr r1 r2 r3 r4 in sy cs us sy id > 1 0 0 36k 184k 0 1 1 0 0 0 0 0 0 0 0 20 100 18 11 4 85 vmstat(1) in this sample doesn't say much except that you about 92 MB free memory, a fairly idle system and not much if any paging I/O. Oh, and probably a load average of 1 or there abouts. If this is the first line that vmstat printed then it's mostly a history of the system since it booted with the "procs" and "memory" being current. The only way to get any real information is to look at successive sample when running vmstat(1) with an argument: vmstat 2 Will print a sample every two seconds. > > The memory fields seem to be fairly consistent around the values shown even > when the load is high. > > This machine has 128 MB of main memory and has ~160 MB ow swap configured. > Pstat -s shows: > > 158361k swap configured > 33471k used (2175k text, 0k smem) > 124888k free, 6999k wasted, 2k missing > This seems consistant with the information displayed by vmstat(1). > My guess would have been that the maximum amount of memory which can get > allocated is 128M (phys. mem) + 160M (swap) less kernel space. Is this right? > No, at least in the current version of Ultrix. Currently the total virtual memory space available is limited to how much page/swap space is configured. Hopefully this will be fixed in a future release. > I have never seen the avm field anywhere close to 128k. Why does the > system swap/page when there seems to be lots of free physical memory? First a comment on "avm". This is ACTIVE virtual memory. Though I've never looked at the kernel sources to see what gets counted as "active" the fact that you 90 something meg free suggests that you shouldn't be paging at all once except to start up a process. Based on the sample vmstat you gave the system doesn't looke it's really paging. Some paging is necessary as part of starting up a process and reading the initialized data, but once that is done it should be "quiet". > > I suspect I'm showing a good deal of ignorance here, but it's true, I am > ignorant. I'm new to both VAX architecture and Unix/Ultrix guts. I'm willing > to rtfm, but would prefer a brief elucidation, with references. The man pages > for vmstat and pstat are more than a little cryptic to me. Don't worry as far as I know there isn't really a FM to read on this. I learned most of what I know by dissecting vmstat, iostat and a program called mon. > > And another thing: I suspect I'm not getting much out of the second > processor. Has this been thrown around before? A while back, when > troubleshooting we turned off the 2nd cpu and didn't notice much of a > hit. Any pointers to understanding how the 2nd proc gets allocated > tasks and how effective it is? The multi-processor version of Ultrix is asymmetric. About the only thing interesting the slave processor(s) can do is run user mode code. There are handful of system calls that can run on a slave, but they aren't that interesting. I haven't looked at the processor scheduling algorithm recently, but think that as long as a process is runnable and there is an idle processor it will run. Remember though that a slave processor can only run user mode code. If your processes make lots of system calls they will only keep the master processor busy dealing with the system calls. A user mode CPU bound processes tend to find an idle processor if there is one and just run to it's heart's content. Multiple CPU bound processes will end up distributing themselves between all the available processors (two being the SUPPORTED max). Iostat(1) should report on the usage of all processors. > > Ignorance is NOT bliss, > Mark. And there is very little in the way of documentation to dispell the ignorance. Just lots of folklore and black magick. Even the distributed programs for collecting the avialable performance information don't tell all they know. Ask your DEC sale rep to track down a program called Monitor (for ULTRIX) and see if he can arrange to get you a copy. Give him my name as a starting place... > > Mark Sausville MIT Media Laboratory -- Alan Rollow alan@nabeth.enet.dec.com