Re: exadata write performance problems

Hi Patrick,

Of course that they will prioritize DBWR/LGWR over everything else, those processes are essential for the functioning of the database. Kerry Osborne even has a blog post which tells you how to do that:

http://kerryosborne.oracle-guy.com/2010/03/increasing-priority-of-lgwr-process-using-_high_priority_processes/

You can do the same thing on linux, using ionice, but only if you are using CFQ scheduler, which is not what Oracle recommends. Oracle recommends "Deadline" scheduler for rotational disks and "noop" for SSD's.

Also, as I have previously said, Exadata is a DW machine, not really optimized for OLTP processing. Waiting on log file sync and parallel control file write means that the underlying database is using a lot of concurrent I/O, which is not what Exadata is made for. Exadata is a DW machine, just like Greenplum and Netezza.

What you really need when it comes to fast writing is a storage that can write really fast. These days, the best results are achieved by using all flash storage. Every major vendor has an all-flash model. If you  want to drive really, really fast, you will not use F-150. If you want to transport a ton of wooden planks from Home Depot to your home, you will not use a Ferrari. It's as simple as that.

Regards

On 2/15/19 7:05 PM, Patrick Jolliffe wrote:
> I'd look seriously to change that, we've had a few problems with
> default setting (BASIC) when hitting high IO load, and it has
> significantly improved by implementing the recommended setting
> (AUTO).  This seems to prioritise critical database writes
> (DBWR/LGWR/???) over others.
> I still don't know why they haven't changed this default if it's not
> the recommended value.
> Starting point would be reading section 6.5.1 of following URL
> https://docs.oracle.com/en/engineered-systems/exadata-database-machine/sagug/exadata-storage-server-iorm.html
> Patrick

-- 
Mladen Gogala
Database Consultant
Tel: (347) 321-1217

--
http://www.freelists.org/webpage/oracle-l