nice 19 doesn't slow down jobs

nice 19 doesn't slow down jobs

Post by Eckhard Rueggebe » Thu, 03 Feb 1994 19:03:37



On a Solaris 1.X machine, if you have two background jobs running with
the same nice value, they get (depending on time already run and memory
requirements) nearly the same CPU percentage. If I renice one to 0 and
one to 8, the one with 8 gets around half of the CPU percentage of the
other. And reniced to 19, it gets hardly any CPU cyles at all.

On a Solaris 2.3 machine, both jobs get the same CPU percentage (near
50% on a otherwise idling machine), NO MATTER how nice they are, even
for 0/19.

How do I change process priority in the normal (non-runtime) class ??

---
Eckhard R"uggeberg

 
 
 

nice 19 doesn't slow down jobs

Post by David P. Ke » Fri, 04 Feb 1994 13:29:19




>On a Solaris 2.3 machine, both jobs get the same CPU percentage (near
>50% on a otherwise idling machine), NO MATTER how nice they are, even
>for 0/19.

>How do I change process priority in the normal (non-runtime) class ??

Nice does set the process priority under Solaris 2, it just doesn't
stay where you set it! :-(.

The Solaris 2 process scheduler is a table-driven state machine. You
can see the transition table by using the command "dispadmin -c TS -g".
Although the tables shipped with Solaris 2.3 are different from 2.2,
both of them are fully connected, meaning that a niced process that
starts out at a particular priority will eventually migrate to the
same priority as other (non-niced) processes.

If you want nice to work as intended, you will have to write and
install your own scheduler table, with disjoint sets of priorities,
so that a process started in a particular set cannot escape from
that set.

It doesn't seem like too much to ask for Sun to supply such a table, but
I guess they have more important things to do.
--

-----------------------------------------------------------------------------
"If it doesn't have an analogy, it isn't Usenet (tm)"

 
 
 

nice 19 doesn't slow down jobs

Post by David Knig » Sun, 06 Feb 1994 22:24:34





>>On a Solaris 2.3 machine, both jobs get the same CPU percentage (near
>>50% on a otherwise idling machine), NO MATTER how nice they are, even
>>for 0/19.

>>How do I change process priority in the normal (non-runtime) class ??

>Nice does set the process priority under Solaris 2, it just doesn't
>stay where you set it! :-(.

>The Solaris 2 process scheduler is a table-driven state machine. You
>can see the transition table by using the command "dispadmin -c TS -g".
>Although the tables shipped with Solaris 2.3 are different from 2.2,
>both of them are fully connected, meaning that a niced process that
>starts out at a particular priority will eventually migrate to the
>same priority as other (non-niced) processes.

>If you want nice to work as intended, you will have to write and
>install your own scheduler table, with disjoint sets of priorities,
>so that a process started in a particular set cannot escape from
>that set.

>It doesn't seem like too much to ask for Sun to supply such a table, but
>I guess they have more important things to do.
>--

So, before I try and reinvent the wheel, has anybody created a disjoint
table for solaris so that all jobs don't migrate to the same priority?
This seems like the way things hould have been set up.
If you have such a table, please post it, and/or mail it to me. Thanks
David
David Knight                          Tel: (518)-442-4204
Department of Atmospheric Science     Fax: (518)-442-4494


 
 
 

1. Nice jobs aren't being nice.

Has the meaning of nice 19 changed from SunOS' definition?
I always found that, under SunOS, my jobs that ran nice 19
would pretty well stop when another job is running.  Under
Solaris 2.3, I'm seeing this from 'top':
--
last pid: 10223;  load averages:  2.06,  2.00,  1.83                   12:19:05
51 processes:  46 sleeping, 4 running, 1 on cpu
Cpu states:  0.0% idle, 95.2% user,  4.8% kernel,  0.0% iowait,  0.0% swap
Memory: 26M real, 1864K free, 33M swap, 116M free swap

  PID USERNAME PRI NICE  SIZE   RES STATE   TIME   WCPU    CPU COMMAND
10207 mjfrazer -22   19 5312K 5096K run     6:13 44.19% 53.28% main
10172 mjfrazer -25    0 3652K 3356K run     6:55 38.19% 41.86% sanchisVLSI
10223 mjfrazer  34    0 1172K  980K cpu     0:06  4.13%  4.76% top
    1 root      34    0  736K  104K sleep  11:46  0.04%  0.00% init
--

Why is my job which is supposedly nice 19 still hogging the CPU?
And, no the job called sanchisVLSI is not ever blocked on IO.

Thanks for any ideas,
mark
--
Mark Frazer    Electrical and Computer Engineering    University of Waterloo

2. DHCP? Automatic Routing info?

3. Nice. Nice nice nice.

4. Mail from command line or bash-script

5. "Nice" or not to "nice" large jobs

6. Async connect code for sockets

7. : Emacs-19.19 can't find term file

8. Best possible outfit to run Linux on...

9. 2.5 slows down on slow ftp's

10. unexpected process scheduler behaviour with nice 19 and fork/vfork

11. 'at' runs job, cron doesn't!

12. Help-102066-09 doesn't like sendmail.cf - mail's down!

13. So what's so nice about nice()?