List Info

Thread: RE: Heavy I/O causes servers to become unusable




RE: Heavy I/O causes servers to become unusable
user name
2007-05-09 15:31:06
I had an issue somewhat related when doing heavy I/O and had
database archive logs running. It ended up being aio issues
such as having large enough values in
/proc/sys/fs/aio-max-nr. I don't remember the details for
sure. But I think there has been a kernel bug related to
aio. But I would see the same behavior with cpu load low but
vmstat showing high I/O on disk. I could even see aio errors
in the alert_log. 

Without making any changes to /proc the 2.6.21 kernel fixed
my problems with this. But if you're running supported SLES
this probably isn't an option for you.

Warren


-----Original Message-----
From: Michael Taylor [mailto:mdtaylor69yahoo.com]
Sent: Wed 5/9/2007 1:24 PM
To: Arun Singh; suse-oraclesuse.com
Subject: Re: [suse-oracle] Heavy I/O causes servers to
become unusable
 
Usually issues like that can be resolved by applying the
latest firmware for all affected systems, (server, storage,
etc) and the latest SLES Service Pack and online updates.

I was able to get a 32 GB DL 585 quad socket dual core up to
a load average of 125 applying large ERP upgrades on SLES-9
when those machines first came out.  After a few firmware
updates and SLES online updates, it was 5 to 10 times faster
than similarly sized Sun boxes.

-Michael

Arun Singh <Arun.Singhnovell.com> wrote: What
is DB version? Are you upto date with SLES9 Service Packs
i.e. SP3 + on-line updates

Is server swapping under heavy load? Please post brief
output of vmstat, iostat and top under I/O load.

-Arun

>>> On 5/9/2007 at 11:58 AM,  wrote:

> We have Oracle databases and applications running on
IBM servers -- SLES8
> on our apps server and SLES9 on our db servers.
> 
> We have our development servers setup with RAID 5.
> 
> We find that when I perform heavy I/O such as copying a
database or
> applying patches or upgrades, the servers become
unusable.  TOP will show
> CPU idle of 35 - 50 % but a load average or 10 - 15. 
This happens on
> internal disks as well as disks on our EMC Clarion so
we suspect it is more
> of a server or OS issue.
> 
> I have been a dba for 12 years and have performed the
same operations on HP
> and Sun Solaris. On Sun and HP, I would see the server
become slow, but not
> unusable.
> 
> Does anyone have any suggestions?
> 
> Angie
> 





--
To unsubscribe, email: suse-oracle-unsubscribesuse.com
For additional commands, email: suse-oracle-helpsuse.com
Please see http://www.suse.com/oracl
e/ before posting



       
---------------------------------
Ahhh...imagining that irresistible "new car"
smell?
 Check outnew cars at Yahoo! Autos.

[1]

about | contact  Other archives ( Real Estate discussion Medical topics )