Renicing and disk usage priority

Chet Nichols III chet.nichols at gmail.com
Fri May 30 20:16:59 UTC 2008


I could be mistaken, but I'd look at it this way: if you nice a  
process to get more cycles, it will be able to use those cycles to  
send new IO requests or receive responses from any previous requests  
sent.. And since you'll be getting more cycles than other processes,  
your wait time should go down.. Even under heavy disk load, your extra  
CPU time should allow you to get stuff to/from the queue more  
frequently.

Then again, your disk won't be performing any faster, it's more like  
you're cutting in line every now and then.

On top of that, IO bound processes get a lot higher (where higher is  
less time) priority than CPU bound processes, so if you nice an IO  
bound process, it will be treated a little more "fair" in a way.

But.. Make sure you play around with it to find the right value. If  
you make it too low, you'll start starving other processes of cycles  
and you'll notice the box hanging a bit, since everything else is  
waiting on your other process checking for IO.

Some of that might not entirely be accurate, but its how I'd go about  
it. If someone has a better answer, feel free to chime in. Thanks!

Chet

-----
Chet Nichols III
Sent from an iPhone


On May 30, 2008, at 9:13 AM, Kristoffer Knigga <Kknigga at arrow-financial.com 
 > wrote:

> Hello, all,
>
> I have a server that's running a bunch of processes that I believe  
> are disk bound.  The server is an 8-way machine running at about 60%  
> idle, and each process averages about 3% of one processor (they are  
> all single threaded).
>
> Now, I have one of these processes that I need to have running at a  
> higher priority when it comes to disk access.  Does renicing a  
> process effect its ability to fight for disk resources, or just  
> processor?
>
> Thanks!
>
> Kris Knigga
>
> -- 
> redhat-list mailing list
> unsubscribe mailto:redhat-list-request at redhat.com?subject=unsubscribe
> https://www.redhat.com/mailman/listinfo/redhat-list




More information about the redhat-list mailing list