Ok, You may want to consider these then. These are for 5.0 on Solaris (but should Be easily transferable to most OS'es and even previous versions). Note that this is generic Guidelines. Depending on memory pressure, application IO profile and a host of other things These may or may not be good tunables. I'd consider them. For databases there is nothing That beats the ODM interface (or the quickIO interface for non-oracle workloads), this is due to the fact that ODM/QIO also alters the filesystem locking behaviour besides the cache behaviour. There are some other tricks we employ as well to accelerate on the ODM/QIO IO path.
Anyway, here is a generic write up on the first set of tunables/mount options I'd consider For vxfs/Vxvm for three workloads; DSS/OLTP/file serving. I would wait with doing anything for The vxtunefs read_pref and nstreams etc before I've analyzed the impact of these tunables since The vxtunefs parameters are autotuned (and autotuned really well). For DMP: Use the balanced path algorithm for A/A type of arrays, consider using a failover group for A/P arrays. Balanced path gives the best performance for almost all workloads. DSS/HPC ========================== VxFS mount options: Log mode= delaylog (best performance and enables data integrity) Logsize= medium (doesn't need to be very much larger than default or keep default) Mincache=direct (turn off in-kernel buffering) Convosync=direct (delay inode updates for writes to files) Tunefs: Read_ahead=1 (traditional vxfs read ahead enabled) Fcl_winterval=6000 (50 minute update interval for FCL) VxVM kernel tuning (/etc/system) Set vxio:vol_maxio=32768 (max IO size = 16MB) OLTP ======================= VxFS mount options: Log mode= delaylog (best performance and enables data integrity) Logsize= medium (doesn't need to be very much larger than default or keep default) Mincache=direct (turn off in-kernel buffering) Convosync=direct (delay inode updates for writes to files) Tunefs: Read_ahead=0 (turn off read ahead) Fcl_winterval=6000 (50 minute update interval for FCL) VxVM kernel tuning (/etc/system) Set vxio:vol_maxio=32768 (max IO size = 16MB) Fileserving, lots of smallish files (like an NFS home directory server) ======================================================================== Log mode= delaylog (best performance and enables data integrity) Logsize= large (Make the log big and use a separate device for the log using volume sets) Tunefs: Read_ahead=0 (turn off read ahead) Fcl_winterval=6000 (50 minute update interval for FCL) kernel tuning (/etc/system) Set vxfs:vxfs_ninode= (this value should be viewed as 1KB units, set to no more than 50% or memory but make it big) Set ncsize= (this should be 80% of vxfs_ninode, the default is derived from maxusers) Other =========================== On some platforms (AIX) you can reduce vx_bc_bufhwm and vx_vmm_buf_count to reduce the size of the FS cache in the kernel, on Solaris this is dynamically done and not directly tuned). You may also want to consider tuning the fsflusher if pid 3 uses a lot of cpu (but that's a sun tunable and not really a vxfs/VxVM tunable). Now, the caveats... These tunables try to optimize the performance by either changing the way memory is consumed or altering IO behavior. Mincacahe, vxfs_ninode, vol_maxio and read_ahead all change memory consumption. Vol_maxio will use some more memory in order to issue larger IO's. Mincache tries to use less IO by doing less caching in the file system. VxFS_ninode tries to use more memory for inode caching to have a larger ncsize cache hit ratio. Log_mode, logsize, fcl_winterval all alter some aspect of write behaviour. Log_mode attempts to delay logging for any non-data updates which strictly doesn't need to be logged in order to conform to posix standards. Logsize should be set so that it never overflows and forces writes to pause due to log overflows. This is hard to monitor btw. But using checkpoints will increase your log device usage and you should consider making the log larger when using checkpoints (btw. Everyone should use checkpoints -it's really good stuff). Fcl_interwal is all about the new file change log. If you update the log at a higher resolution than needed then we put unneeded write pressure against the underlying device (and the filesystem at large. Set it to a high enough write interval to be useful for the monitoring requirements you have. 50 minutes or so will give you a really high resolution which is useful for day to day monitoring but not useful for intra-day monitoring (at least in my mind). Best regards, Par -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of [EMAIL PROTECTED] Sent: Thursday, October 26, 2006 2:59 AM To: veritas-vx@mailman.eng.auburn.edu Subject: Re: [Veritas-vx] Recommended mount options for VxFS OLTP (Oracle) on Solaris 10 On Tue, Oct 24, 2006 at 06:20:03AM -0700, Par Botes wrote: > What's the workload? Transactional or DSS style queries in the database? > It doesn't change the recommended mount options but it changes some of > the VxVM recommendations. > > Platform? There are come kernel tunables which could be modified as > well. > > -Par > > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of > [EMAIL PROTECTED] > Sent: Monday, October 23, 2006 11:21 PM > To: veritas-vx@mailman.eng.auburn.edu > Subject: Re: [Veritas-vx] Recommended mount options for VxFS > > Thanks Par for the advice but, as far as I know, to have ODM we need > to buy it. > I am asking for mount options in typical VxFS+VxVM environment. > > przemol > > On Mon, Oct 23, 2006 at 10:16:02AM -0700, Par Botes wrote: > > No mount options needed if you use the ODM option of VxFS (or if you > > use the older QuickIO interface). I strongly recommend anyone who > > has oracle to use ODM (or QIO in the past), it makes a difference. > > > > If you do a lot of checkpoints you may want to increase the log size > > and put it on a different device as well. > > > > -Par > > > > > > -----Original Message----- > > From: [EMAIL PROTECTED] > > [mailto:[EMAIL PROTECTED] On Behalf Of > > [EMAIL PROTECTED] > > Sent: Monday, October 23, 2006 3:56 AM > > To: veritas-vx@mailman.eng.auburn.edu > > Subject: [Veritas-vx] Recommended mount options for VxFS > > > > Hi all, > > > > can you share some information what options (convosync,mincache, > > other) you use mounting VxFS for Oracle database ? Do you treat the > > same redo logs, database files and archive logs or mount it with > > different options ? > > > > Regards > > przemol > > > > -------------------------------------------------------------------- > > -- PS. >>> http://link.interia.pl/f19a6 > > > > _______________________________________________ > > Veritas-vx maillist - Veritas-vx@mailman.eng.auburn.edu > > http://mailman.eng.auburn.edu/mailman/listinfo/veritas-vx > > ---------------------------------------------------------------------- > Jestes kierowca? To poczytaj! >>> http://link.interia.pl/f199e > > _______________________________________________ > Veritas-vx maillist - Veritas-vx@mailman.eng.auburn.edu > http://mailman.eng.auburn.edu/mailman/listinfo/veritas-vx ---------------------------------------------------------------------- PS. >>> http://link.interia.pl/f19a6 _______________________________________________ Veritas-vx maillist - Veritas-vx@mailman.eng.auburn.edu http://mailman.eng.auburn.edu/mailman/listinfo/veritas-vx _______________________________________________ Veritas-vx maillist - Veritas-vx@mailman.eng.auburn.edu http://mailman.eng.auburn.edu/mailman/listinfo/veritas-vx
