|
Home > Archive > Unix administration > April 2004 > how to share cpu busyness across all the cpus
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
how to share cpu busyness across all the cpus
|
|
| yls177 2004-03-26, 11:46 am |
| i do a sar -P ALL -u 1 10
find that one or two cpus are 100% occupied while the rest are all idling..
any way to auto load balancing for them?
| |
| Michael Vilain 2004-03-26, 11:46 am |
| In article <c06e4d68.0403252029.7731059@posting.google.com>,
yls177@hotmail.com (yls177) wrote:
> i do a sar -P ALL -u 1 10
>
> find that one or two cpus are 100% occupied while the rest are all idling..
>
> any way to auto load balancing for them?
You didn't specify what OS and version you'er running, so I'll reply in
general. SMP systems _do_ load balance across multiple processors. You
may not have enough of a load to use all the CPUs, especially if your
workload isn't CPU intensive. You may be doing a lot of IO, where
having multiple CPUs won't help you nor can you load balance this. Have
you analyzed your performance and do you know what your bottleneck is
(it's either CPU, IO or memory)?
Having multiple CPUs only benefits you when you have enough compute
bound processes to use them. Also, are your compute-bound processes
multi-threaded? If not, you won't get any gain from running them on
multiple CPU systems. Buy a bigger, faster system instead or rewrite
the application to be multi-threaded.
--
DeeDee, don't press that button! DeeDee! NO! Dee...
| |
| Alan Coopersmith 2004-03-28, 1:35 pm |
| yls177@hotmail.com (yls177) writes in comp.unix.admin:
|i do a sar -P ALL -u 1 10
|find that one or two cpus are 100% occupied while the rest are all idling..
|any way to auto load balancing for them?
If you only have one or two processes needing to use a lot of CPU, then
forcing them to move among CPU's will simply slow them down with no
benefit, due to all the overhead of moving their state from CPU to CPU,
and the loss of all the data they've cached in the local caches for the
CPU.
--
________________________________________
________________________________
Alan Coopersmith alanc@alum.calberkeley.org
http://www.CSUA.Berkeley.EDU/~alanc/ aka: Alan.Coopersmith@Sun.COM
Working for, but definitely not speaking for, Sun Microsystems, Inc.
| |
| yls177 2004-03-29, 12:35 am |
| Alan Coopersmith <alanc@alum.calberkeley.org> wrote in message news:<c474p3$20so$1@agate.berkeley.edu>...
> yls177@hotmail.com (yls177) writes in comp.unix.admin:
> |i do a sar -P ALL -u 1 10
> |find that one or two cpus are 100% occupied while the rest are all idling..
> |any way to auto load balancing for them?
>
> If you only have one or two processes needing to use a lot of CPU, then
> forcing them to move among CPU's will simply slow them down with no
> benefit, due to all the overhead of moving their state from CPU to CPU,
> and the loss of all the data they've cached in the local caches for the
> CPU.
from what i could understood from your posts, is that its best to
leave the sceduling of the cpus to the os itself. whether, its one or
2 busy cpus, the os will be in the best state to know what is the best
for the system.
right?
| |
| Jay Lessert 2004-03-29, 5:37 pm |
| yls177@hotmail.com (yls177) wrote in message news:<c06e4d68.0403252029.7731059@posting.google.com>...
> i do a sar -P ALL -u 1 10
>
> find that one or two cpus are 100% occupied while the rest are all idling..
>
> any way to auto load balancing for them?
Most likely, you have two single-threaded processes, and your
system is doing exactly what you're telling it to do.
-Jay-
| |
|
| jayl-news@accelerant.net (Jay Lessert) wrote in message news:<7109f92b.0403291403.57671b97@posting.google.com>...
> yls177@hotmail.com (yls177) wrote in message news:<c06e4d68.0403252029.7731059@posting.google.com>...
>
> Most likely, you have two single-threaded processes, and your
> system is doing exactly what you're telling it to do.
>
> -Jay-
We used to have this problem on an oracle db
So the software company rewrote it to be multithreading
Now we have a quicker system that core dumps and has spinning out processes
But it is quicker ;)
| |
| yls177 2004-04-03, 5:35 am |
| jedbeer@hotmail.com (Ed) wrote in message news:<dc34c090.0404010626.2ead41fe@posting.google.com>...
> jayl-news@accelerant.net (Jay Lessert) wrote in message news:<7109f92b.0403291403.57671b97@posting.google.com>...
>
> We used to have this problem on an oracle db
>
> So the software company rewrote it to be multithreading
>
> Now we have a quicker system that core dumps and has spinning out processes
>
> But it is quicker ;)
i am running sap on these box.... so any thing that i can do to
improve the situation. thanks
|
|
|
|
|