[net.unix-wizards] run queues on Vaxen running Unix

rcj@burl.UUCP (R. Curtis Jackson) (05/30/85)

Just curious to know the largest runques observed on Vaxen.  I am running
on several different Vax 11/780s all running SVR2.  One machine always
has the largest runques; I take samples via 'sar -q 3 20'; it takes an
average every 3 seconds for a minute (20 times) and gives an average
figure.  The worst I had seen before this machine was an average of 17.

Yesterday I did one of these and the average was 40.3, with a high
runque of 46.8!!!

Anyone seen anything worse than this?
-- 

The MAD Programmer -- 919-228-3313 (Cornet 291)
alias: Curtis Jackson	...![ ihnp4 ulysses cbosgd mgnetp ]!burl!rcj
			...![ ihnp4 cbosgd akgua masscomp ]!clyde!rcj

guy@sun.uucp (Guy Harris) (06/01/85)

> Just curious to know the largest runques observed on Vaxen.  I am running
> on several different Vax 11/780s all running SVR2.  One machine always
> has the largest runques; I take samples via 'sar -q 3 20'; it takes an
> average every 3 seconds for a minute (20 times) and gives an average
> figure.  The worst I had seen before this machine was an average of 17.

For the benefit of BSD people, "runque" ~= "instantaneous load average".
S5R2 keeps two counters; one for the number of loaded processes on the run
queue and one for the number of swapped processes on the run queue.  The
4.2BSD "load average" also includes sleeped or stopped processes sleeping at
a "nonpositive" priority (<= PZERO).  The "total" fields "t_rq" and "t_sq"
correspond to the "runque" and "swpque" values.

	Guy Harris

kre@munnari.OZ (Robert Elz) (06/01/85)

In article <716@burl.UUCP>, rcj@burl.UUCP (R. Curtis Jackson) writes:
> Just curious to know the largest runques observed on Vaxen...
> 
> Yesterday I did one of these and the average was 40.3, with a high
> runque of 46.8!!!
> 
> Anyone seen anything worse than this?

  3:36pm  up  2:09,  37 users,  load average: 108.53, 104.78, 98.49

That's output from 4.1 something's uptime (I forget which version we
were running then, it might even have been very early 4.2), in late
October 83.  As you can see (if you understand the format) the load
was still rising.  It peaked at something over 110...

This is from munnari, a 780 with 4Mb, fpa, etc (but not "set clock fast")

The system had only been up 2 hours, as just before then the load
was SO bad that we had to reboot to get things back into a stable
state.

Robert Elz					seismo!munnari!kre

(ps: in case you're wondering, most of those users would have been
running ingres - there would also have been a bunch of f77 compiling)

muller@sdcc3.UUCP (Keith Muller) (06/04/85)

I wish I had saved the line, but in Feb 1983 sdcc3 running 4.1c(?) had
a load of over 125. (Sdcc3 is a 780 with 4 megs 64 ports, ethernet and
had about 2500 active student users). The machine made the local TV news
that week as students were camping out in the terminal rooms with tents
waiting for a terminal port. This load probably went a lot higher, but it
took 50 minutes for that uptime to respond and it was 3am in the morning
and I wasn't going to try another one.
I know if I turned off the load control system the runq would easily top 125
on any one of the student machines here (it is the last week of the quarter
here).
	Keith Muller
	University of California

chris@umcp-cs.UUCP (Chris Torek) (06/05/85)

> This load probably went a lot higher, but it took 50 minutes for
> that uptime to respond and it was 3am in the morning and I wasn't
> going to try another one.

Another argument for control-T!  :-)

  load 3.52 3.27 3.37, pid 13179, %cpu 1.27, 28k of 117k, kbd wait

(Or if you prefer: after ``stty -load5 -load15 uptime children''

  up 12h 49m, load 3.52, pid 13179 children 1, %cpu 0.68, 28k of 117k
  kbd wait

or I could turn off pid or size or whatever.)
-- 
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 4251)
UUCP:	seismo!umcp-cs!chris
CSNet:	chris@umcp-cs		ARPA:	chris@maryland

kay@warwick.UUCP (Kay Dekker) (06/07/85)

In article <2898@sdcc3.UUCP> muller@sdcc3.UUCP (Keith Muller) writes:
>The machine made the local TV news
>that week as students were camping out in the terminal rooms with tents
>waiting for a terminal port.

Camping out in the terminal rooms with *tents*?  Why (he asked, surprised)
do you need a tent to camp in a terminal room?  Or is the weather kind enough
in California to permit you to have open-roofed terminal rooms?

								Kay.



-- 
"In a world without rational structure, even the most bizarre events must
eventually take place."   -- Philip Avalon, "On the Resurrection of Reagan"
			
			... mcvax!ukc!warwick!flame!kay

mjl@ritcv.UUCP (06/09/85)

In article <6299@umcp-cs.UUCP> chris@umcp-cs.UUCP (Chris Torek) writes:
>> This load probably went a lot higher, but it took 50 minutes for
>> that uptime to respond and it was 3am in the morning and I wasn't
>> going to try another one.
>
>Another argument for control-T!  :-)

Control-T is nice, but when the load average is 125, it's just quicker
at telling you how miserable things really are :-(
-- 
Mike Lutz	Rochester Institute of Technology, Rochester NY
UUCP:		{allegra,seismo}!rochester!ritcv!mjl
CSNET:		mjl%rit@csnet-relay.ARPA