Linux 10GBE Tuning?
- Thread starter Dogers
- Start date
Its just TCP tuning; so things like: http://fasterdata.es.net/host-tuning/linux/
Soldato
- Joined
- 18 Oct 2002
- Posts
- 4,898
What do you mean by redo logs "timing out"? The database can only process DML as fast as it can write the redo, so if it's really slow it can bring the database to a crawl but I've never heard of redo timing out. The only timeout issue with redo I can think of is if your redo logs fill up before Oracle has had chance to archive them - that will stop DML dead and suggests you need more redo log groups adding and/or the size of the log files increasing.
If the speed of redo writing is limiting your DB performance and the hardware is good, then the solution is to go to RAC.
Are you using Direct NFS on Oracle? If not, you probably should be. Also check your mount option are as recommended by Oracle in the install guide.
Are you using multiple NFS mounts? I've not used Oracle on Linux, but on Solaris using multiple mount points for the different file types improves performance, even if they are ultimately held on the same physical storage. As a minimum split your datafiles, online redo logs and archive log locations to separate mounts.
We use Oracle on Solaris on 2x 10GbE LACP trunk for storage and 4x 1GbE out to the users. We've done a fair bit of tuning but it's not made vast leaps in performance. Direct NFS gave us the biggest jump.
If the speed of redo writing is limiting your DB performance and the hardware is good, then the solution is to go to RAC.
Are you using Direct NFS on Oracle? If not, you probably should be. Also check your mount option are as recommended by Oracle in the install guide.
Are you using multiple NFS mounts? I've not used Oracle on Linux, but on Solaris using multiple mount points for the different file types improves performance, even if they are ultimately held on the same physical storage. As a minimum split your datafiles, online redo logs and archive log locations to separate mounts.
We use Oracle on Solaris on 2x 10GbE LACP trunk for storage and 4x 1GbE out to the users. We've done a fair bit of tuning but it's not made vast leaps in performance. Direct NFS gave us the biggest jump.
I'm not entirely sure of the whole story yet - I'm coming at it from the storage view. They're using DBVisit and say that when it comes to run the redo/catchup/whatever, the previous one is still running. On the old DAS shelf it wasn't..
From looking at SAR logs on the host, we're not doing more than 300Mbit at peak, so thought we'd see what's out there for tweaking that. They're not using dNFS for some reason (I seem to recall it's the version of Oracle in use - 11Gr1 is/was pants for dNFS IIRC?) so that could be another reason..
From looking at SAR logs on the host, we're not doing more than 300Mbit at peak, so thought we'd see what's out there for tweaking that. They're not using dNFS for some reason (I seem to recall it's the version of Oracle in use - 11Gr1 is/was pants for dNFS IIRC?) so that could be another reason..