Re: ioWait brown out

Aaron Birenboim <[email protected]> Fri, 04 Feb 2011 11:24:22 -0700
Newsgroups gmane.org.user-groups.nmlug
Message-ID <[email protected]>
On 02/04/2011 11:02 AM, Matthew Bowie wrote:
> Are they created one at a time, or are you creating and writing to all
> of them simultaneously?  If the latter, it's possible that the seek
> times are what are killing you.

one at a time, serially.

>  Are you doing anything that would
> force writes (like fsync) and prevent the OS caching from working?

I have put sync() calls in there, and it made little difference.
Instead of my program exit in 2-3 minutes, and lock the system
for the next 10 minutes, with sync in there it might take
10 minutes to run, and lock the system for 5 minutes afterwards.
I would think that fsync() should help....  If I could throttle
my process down to wait for 50MB/sec disk writes, I'd be thrilled.
It is more like 1MB/sec now... and my machine semi-locks up while
it is doing this, and for 10 minutes after I'm done.


> What kind of write rates do you see writing to a single file using dd?

 > sudo sync;dd if=/dev/zero of=t bs=1024 count=100000;sudo sync
100000+0 records in
100000+0 records out
102400000 bytes (102 MB) copied, 0.547869 s, 187 MB/s

Printed output almost immediately, but took 50 sec for
the second sudo sync to complete!
Who else has this?

Most of my systems are Debian or Ubuntu.

On my fileserver, I got:

sudo sync;dd if=/dev/zero of=t bs=1024 count=100000;sudo sync
[sudo] password for aaron:
100000+0 records in
100000+0 records out
102400000 bytes (102 MB) copied, 0.238117 s, 430 MB/s

returned almost immediately.  hmmmmm.....

What might be different about my workstations?
(I have tried three of them)

                     aaron
_______________________________________________
NMLUG mailing list
[email protected]
http://lists.b9.com/cgi-bin/mailman/listinfo/nmlug