darren@bacchus (Darren Friedlein) (05/20/88)
I have a question that I'm hoping someone can answer... I run phdaemon on
my machine and when memory has to be swapped out to disk, phdaemon fails,
leaves me a message under the [!!] icon and I re-start it. No big
problem. The last two times, however, the [!!] icon flashed on and then
the whole status bar right of "DATA 2:" went to dots. I assume that
this means that wmgr is failing when it is swapped out too. This is
pretty damn agrivating -- especially since it never happened before.
I have a UNIXpc w/ a 40M drive, 1M on the MB and v3.5(.1.4) of the OS
(note: not 3.51). I'm using the new wmgr that moves between windows
with the <Resume> key, but I've been using that all along.
Anyone got any ideas???
-darren
____
/ \
| Rt 4, Box 416, Durham, NC 27703
_____|_____ Darren G. Friedlein data (bacchus) : 919/596-7746
/ | \ voice : 919/596-9492
( | )
\____/ __/ {mcnc|icus|ethos|gladys|bakerst}!bacchus!darrenford@elgar.UUCP (Ford Prefect ) (05/23/88)
In article <449@bacchus> darren@bacchus (Darren Friedlein) writes: >I have a question that I'm hoping someone can answer... I run phdaemon on >my machine and when memory has to be swapped out to disk, phdaemon fails, >leaves me a message under the [!!] icon and I re-start it. What is the message? How do you know it has anything to do with swapping (and do you really mean swapped, as in a "0" Flags entry in ps -fl, or just normal VM paging)? > No big >problem. The last two times, however, the [!!] icon flashed on and then >the whole status bar right of "DATA 2:" went to dots. I have seen wmgr (on 3.51) dump core when I do some bizarre things with s_Suspd/s_Rsume/Suspd while windows are being created/deleted. Maybe it's that? But what you describe means that smgr went bye- bye as well. Look for a core file if in happens again. > I assume that >this means that wmgr is failing when it is swapped out too. It is very difficult for a program to find out whether it has been swapped out. Are you suspecting a system-wide problem with swap errors? If you ever run a program and immediately see "Killed" instead, that probably means you are having I/O errors on the swap device. If you find a core file of wmgr, however, it's probably just a bug in wmgr which almost certainly has nothing to do with paging. On a standard system, wmgr's current directory is /etc/lddrv, so that is the place to look for the core. -=] Ford [=- "Once there were parking lots, (In Real Life: Mike Ditto) now it's a peaceful oasis. ford%kenobi@crash.CTS.COM This was a Pizza Hut, ...!sdcsvax!crash!kenobi!ford now it's all covered with daisies." -- Talking Heads
dga@killer.UUCP (David Aldrich) (05/24/88)
in reference to one deamon killing off another on the UNIX-PC each of the programs ph smgr and wmgr control a few keys on the keyboard and a space on the top line of the unixpc if any of these programs dies for any reason at all like odd instructions being sent to them etc. thier area on the top line goes blank. The 'w' Icon is owned by wmgr and is in /dev/w4 The center are is owned by smgr and is /dev/w5 and the left is /dev/w3 and owned by ph the windows on the UNIX PC are a neat idea and if there is some interest I have made up some programs to use these from the shell without hitting suspd or the like. also if you are tired of a non full 80x24 shell window from the office I have a program to resize your current window to full screen and some routines to make up arbitrary sizes and places for windows if you have a particular request and a compiler I can mail the source with a short explanation or post it if there is interest. I am killer!dga
lenny@icus.UUCP (Lenny Tropiano) (05/24/88)
In article <449@bacchus> darren@bacchus (Darren Friedlein) writes: |> |>I have a question that I'm hoping someone can answer... I run phdaemon on |>my machine and when memory has to be swapped out to disk, phdaemon fails, |>leaves me a message under the [!!] icon and I re-start it. No big |>problem. The last two times, however, the [!!] icon flashed on and then |>the whole status bar right of "DATA 2:" went to dots. ... Well I wrote phdaemon and never really had the ambition to stick the code in it to open /dev/swap and locate the user's process block if it got swapped out. Unfortunately this is a very hard thing to *TEST*. How do you make something run out of swap space if you have enough allocated? I suggest you (yes, reformat) and increase your default swap space. Running out of swap space isn't too good, it's a sign that your machine isn't configured correctly for the amount of memory you have and how intensively you use this memory. -Lenny -- US MAIL : Lenny Tropiano, ICUS Computer Group IIIII CCC U U SSS PO Box 1 I C U U S Islip Terrace, New York 11752 I C U U SS PHONE : (516) 968-8576 [H] (516) 582-5525 [W] I C U U S TELEX : 154232428 [ICUS] IIIII CCC UUU SSS AT&T MAIL: ...attmail!icus!lenny UUCP : ...{mtune, ihnp4, boulder, talcott, sbcs, bc-cis}!icus!lenny
darren@bacchus (Darren Friedlein) (05/24/88)
In article <153@elgar.UUCP> ford@kenobi.UUCP (Mike Ditto) writes: >In article <449@bacchus> darren@bacchus (Darren Friedlein) writes: >>I have a question that I'm hoping someone can answer... I run phdaemon on >>my machine and when memory has to be swapped out to disk, phdaemon fails, >>leaves me a message under the [!!] icon and I re-start it. > >What is the message? How do you know it has anything to do with >swapping (and do you really mean swapped, as in a "0" Flags entry in >ps -fl, or just normal VM paging)? When smgr (found out I had the wrong mgr) displayed the [!!] icon before, the message I got was that phdaemon died because SOMETHING was swapped out, either phdaemon or a program it was monitoring. This time, smgr quit completely right after the [!!] icon appeared. From ps, I could see that the system load was real heavy. This doesn't assure that swapping was the cause, but that would be my best guess. What is the difference between a process being swapped out and normal VM paging? I thought the UNIXpc could only support 4M or virtual memory, but when I formatted the drive, it reserved 6M of space. > -=] Ford [=- > >"Once there were parking lots, (In Real Life: Mike Ditto) >now it's a peaceful oasis. ford%kenobi@crash.CTS.COM >This was a Pizza Hut, ...!sdcsvax!crash!kenobi!ford >now it's all covered with daisies." -- Talking Heads ____ / \ | Rt 4, Box 416, Durham, NC 27703 _____|_____ Darren G. Friedlein data (bacchus) : 919/596-7746 / | \ voice : 919/596-9492 ( | ) \____/ __/ {mcnc|icus|ethos|gladys|bakerst}!bacchus!darren
jbm@uncle.UUCP (John B. Milton) (05/25/88)
In article <449@bacchus> darren@bacchus (Darren Friedlein) writes: ... >problem. The last two times, however, the [!!] icon flashed on and then >the whole status bar right of "DATA 2:" went to dots. I assume that >this means that wmgr is failing when it is swapped out too. This is The phone manager (ph) is the left half of the status line (w3), the window manager (wmgr) is just the upper right corner (w4), and the status manager (smgr), which is really an expanded cron (w5) is most of the right half. When smgr (cron), ph or wmgr die, you will find a core file in /etc/lddrv. I am personally very suspect of smgr and the way it intereacts with the window driver. Partial ps -f showiing window devcies open: root 83 1 3 May 23 w4 0:01 wmgr root 78 1 3 May 23 w3 0:33 ph root 87 1 3 May 23 w5 4:45 smgr John -- John Bly Milton IV, jbm@uncle.UUCP, {ihnp4|osu-cis}!n8emr!uncle!jbm home: (614) 294-4823, work: (614) 459-7641; talk to me about fractals
dga@killer.UUCP (David Aldrich) (05/26/88)
Actually I was thinking about some of the problems with the 'ph' program. You may try <if you don't use it of course> taking it out of /etc/rc so that it will never be spawned in the first place then it of course could not interfere with you phone operation. I have killed it before then used cu etc with no noticible difference in opertion. If anyone witha uucp that has trouble does this and finds that it helps then we will have narrowed the problem down some.
ford@crash.cts.com (Michael Ditto) (05/27/88)
In article <459@bacchus> darren@bacchus.UUCP (Darren Friedlein) writes: >In article <153@elgar.UUCP> ford@kenobi.UUCP (Mike Ditto) writes: >>In article <449@bacchus> darren@bacchus (Darren Friedlein) writes: >>>I have a question that I'm hoping someone can answer... I run phdaemon on >>>my machine and when memory has to be swapped out to disk, phdaemon fails, >>>leaves me a message under the [!!] icon and I re-start it. >> >>What is the message? How do you know it has anything to do with >>swapping (and do you really mean swapped, as in a "0" Flags entry in >>ps -fl, or just normal VM paging)? > >When smgr (found out I had the wrong mgr) displayed the [!!] icon before, >the message I got was that phdaemon died because SOMETHING was swapped >out, either phdaemon or a program it was monitoring. Ahh... I'm beginning to understand the setup... I had no idea that phdaemon was such a sneaky program as to go peeking at other processes' user information. So what happened is phdaemon tried to look at the user structure for something (probably the /etc/ph process) and found that it was not in memory. Phdaemon's author (isn't that you, Lenny?) didn't provide for this situation and had the program exit instead (probably based on the first version of "fuser", which did the same thing). > This time, smgr >quit completely right after the [!!] icon appeared. From ps, I could >see that the system load was real heavy. This doesn't assure that >swapping was the cause, but that would be my best guess. I would guess that that was not directrly related, unless phdaemon is sending some weird stuff to the smgr. Perhaps you ran out of swap space. >What is the difference between a process being swapped out and normal >VM paging? I thought the UNIXpc could only support 4M or virtual memory, >but when I formatted the drive, it reserved 6M of space. Normally, individual pages of memory are moved to the swap device when they aren't needed, and brought back in individually when something tries to use them. Processes are "swapped out" when the system gets desparate for space. A swapped process is completely stored on disk and can not execute at all until it is (at least partially) swapped in. If a process is swapped out, bit zero of the "F" field of the ps display will be zero. The Unix PC supports 4M of virtual memory PER PROCESS, with a total amount dependant on your physical memory + swap space. By the way, Unix completely hides the whole questions of swapping and paging from user programs. It's only in the rare case of a weird program like ps, fuser, or (apparrently) phdaemon that it comes up. That's why I was really surprised to hear someone claim that they were getting errors because of swapping. -- Mike Ditto -=] Ford [=- P.O. Box 1721 ford%kenobi@crash.CTS.COM Bonita, CA 92002 ford@crash.CTS.COM
jbm@uncle.UUCP (John B. Milton) (05/29/88)
In article <3030@crash.cts.com> ford@crash.CTS.COM (Michael Ditto) writes: >In article <459@bacchus> darren@bacchus.UUCP (Darren Friedlein) writes: >>In article <153@elgar.UUCP> ford@kenobi.UUCP (Mike Ditto) writes: >>>In article <449@bacchus> darren@bacchus (Darren Friedlein) writes: ... >The Unix PC supports 4M of virtual memory PER PROCESS, with a total amount >dependant on your physical memory + swap space. I read somewhere in the docs that the UNIXpc actually limits each process to 2.5M virtual. It could not be as high as 4.0M because the kernel ALWAYS takes up some 340k+. Then there are loadable drivers, etc. If user processes were allowed to get near 4M, they could easily force the machine into constant thrash and eventual crash. I just whipped up a quicky, and here's what I got (2.5M=2609152) max allocate ~= 2535424 main() { int first,last,old; first=sbrk(0); while ((old=sbrk(4096))!=-1) last=old; printf("max allocate ~= %d\n",last-first); } John -- John Bly Milton IV, jbm@uncle.UUCP, {ihnp4|osu-cis}!n8emr!uncle!jbm home: (614) 294-4823, work: (614) 459-7641; talk to me about fractals
ford@elgar.UUCP (Ford Prefect ) (06/01/88)
In article <271@uncle.UUCP> jbm@uncle.UUCP (John B. Milton) writes: >In article <3030@crash.cts.com> ford@crash.CTS.COM (Michael Ditto) writes: >... >>The Unix PC supports 4M of virtual memory PER PROCESS, with a total amount >>dependant on your physical memory + swap space. > >I read somewhere in the docs that the UNIXpc actually limits each process to >2.5M virtual. It could not be as high as 4.0M because the kernel ALWAYS takes >up some 340k+. Then there are loadable drivers, etc. If user processes were >allowed to get near 4M, they could easily force the machine into constant >thrash and eventual crash. Yes, 2.5M is the maximum "normal" memory in one process. The four meg address space is used up by: .5M of kernel, 2.5M of user code/stack/data, .5M for the shared library, and .5M for shared memory. The 340k figure for the kernel is the physical memory for the plain kernel; loadable drivers will use more physical memory, but no user virtual memory because they first fill up the rest of the kernel's .5M and then start in the shared library range. I don't think you can crash a machine by making it thrash for memory, and besides, that isn't caused by a process that allocates lots of memory unless it is actually referencing lots of pages. You can make things awfully slow though. >I just whipped up a quicky, and here's what I got (2.5M=2609152) > >max allocate ~= 2535424 > >main() >{ > int first,last,old; > > first=sbrk(0); > while ((old=sbrk(4096))!=-1) > last=old; > printf("max allocate ~= %d\n",last-first); >} If you add to that 2535424 the memory taken up by the program itself and its data and stack, you'd probably be within one page of the 2M limit. If you need more, you could create half a meg of shared memory! -=] Ford [=- "Once there were parking lots, (In Real Life: Mike Ditto) now it's a peaceful oasis. ford%kenobi@crash.CTS.COM This was a Pizza Hut, ...!sdcsvax!crash!kenobi!ford now it's all covered with daisies." -- Talking Heads