fortinp@bcars223.bnr.ca (Pierre Fortin) (03/19/90)
While the problem I am about to describe occurs on ST 1.1, has anyone
seen it on the PC version and/or any of the recent upgraded versions?
Using ST MINIX 1.1, I began the process of upgrading from the net postings.
Once I had all the original 1.5.0 postings on my HD, I figured this MINIX
sutff is multi-tasking, right? All the postings were on my disk as: 00,
01, 02, ..., 73. So I started typing in the following commands (I know,
I could use a shell script):
uudecode 01 &
uudecode 02 &
uudecode 03 &
uudecode 04 &
etc.
Then things started to get REAL SLOW. Before anyone jumps to conclusions
about "Yeah, but you're running all those processes...", here's something
to ponder...
Using the "ps" hot-key (Alt/Ctrl/F1), I noticed that usually one or two of
these processes were using inordinate amounts of CPU time. During the
01-73 posting uudecodes, I found that all will eventually go to completion.
Once one of these processes is in this state (it may be the ONLY process
still running), it does I/O at the rate of about 1 per 8-10 seconds. Also,
the CPU times which are usually in the low three-digit numbers climb to
16000+!!! units in the "sys" column.
I also noted that when two processes are in this state simultaneously,
they will have alternating "flag" of 0 or 8, and one or the other will have
FS in the last column before the "command"; never both at the same time.
Killing the errant processes and restarting them usually runs fine.
NOTE: this problem is highly repeatable; entering enough commands also
produces the "try again" message. The problem is NOT related with
"uudecode", since I got the same thing (after re-booting) when
running the resultant shells via "sh bawk_00", "sh bawk_01", etc.
Is this a known problem? Has it been fixed in the recent postings?
Pierre Fortin
fortinp@bnr.camckee@hpfcdc.HP.COM (Bret McKee) (03/20/90)
>Using ST MINIX 1.1, I began the process of upgrading from the net postings. >Once I had all the original 1.5.0 postings on my HD, I figured this MINIX >sutff is multi-tasking, right? All the postings were on my disk as: 00, >01, 02, ..., 73. So I started typing in the following commands (I know, >I could use a shell script): > > uudecode 01 & > uudecode 02 & > uudecode 03 & > uudecode 04 & > etc. > >Then things started to get REAL SLOW. Before anyone jumps to conclusions >about "Yeah, but you're running all those processes...", here's something >to ponder... >Using the "ps" hot-key (Alt/Ctrl/F1), I noticed that usually one or two of >these processes were using inordinate amounts of CPU time. During the >01-73 posting uudecodes, I found that all will eventually go to completion. >Once one of these processes is in this state (it may be the ONLY process >still running), it does I/O at the rate of about 1 per 8-10 seconds. Also, >the CPU times which are usually in the low three-digit numbers climb to >16000+!!! units in the "sys" column. On the PC 1.1, there is a problem with the CPU accounting. It does not handle lots of things correctly. Also, the single threaded filesystem and device drivers will really kill you running those uudecodes... Bret --- Bret Mckee Hewlett Packard HP-UX Development Lab Phone:(303)229-6116 email: mckee@hpmckee or mckee%hpmckee@hplabs.hp.com Copyright (c) Bret Mckee 1990. All Rights Reserved. Of course, these are just my opinions...
fortinp@bcars223.bnr.ca (Pierre Fortin) (03/24/90)
In article <6780003@hpfcdc.HP.COM>, mckee@hpfcdc.HP.COM (Bret McKee) writes: [from my original posting] > >Using the "ps" hot-key (Alt/Ctrl/F1), I noticed that usually one or two of > >these processes were using inordinate amounts of CPU time. During the > >01-73 posting uudecodes, I found that all will eventually go to completion. > >Once one of these processes is in this state (it may be the ONLY process > >still running), it does I/O at the rate of about 1 per 8-10 seconds. Also, > >the CPU times which are usually in the low three-digit numbers climb to > >16000+!!! units in the "sys" column. > > On the PC 1.1, there is a problem with the CPU accounting. It does not > handle lots of things correctly. Also, the single threaded filesystem and > device drivers will really kill you running those uudecodes... Maybe on the PC 1.1, but my problem is on ST 1.1 which is equivalent to PC 1.3. I could understand if all these processes were still running, but all those which were unaffected go to completion, then I'm left with one or two processes (I've had the same problem with other than uudecode) which seem to be in a sort of time warp (with the meter running). Everything else seems to be fine, just the process(es) which are in this state are affected. Why, I can even fire up another process which, if it is not affected, will zip right through doing its I/O at top speed while the affected one(s) is still chewing up CPU time and doing extremely long waits between its I/Os. It's kind of strange to see something like a small uudecode taking minutes while a much longer one starts and finishes in seconds. > > Bret > > --- > Bret Mckee > Pierre Fortin fortinp@bnr.ca > > Of course, these are just my opinions... ...and of course, these are just my observations... :^)