Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Latch Contention

Re: Latch Contention

From: <muller.brenda_at_primestar.tci.com>
Date: 1997/02/25
Message-ID: <856887514.19314@dejanews.com>#1/1

In article <Pine.HPP.3.91.970221145417.1450A-100000-100000-100000-100000-100000-100000_at_s.mincom.oz.au>,   Margaret Pitts <mpitts_at_saturn.mincom.oz.au> wrote:
>
> > In regards to LATCH contention:
> > I can't remember what they told us in "Oracle Tune and Troubleshooting
> > Class" about multiple CPU's and the init.ora parameter
> > "LOG_SIMULTANEOUS_COPIES"???? for copy latch. I can't remember what the
> > instructor said that when adding a second or more CPU's that a latch
> > parameter could be bumped up. Or was it spin_count?
> > We have 2 250MHz Alpha CPU's, just added the second a couple months ago.
> > (DEC UNIX 3.2)
> > I just want to be optimal when we add another 50+ financial users.
> >
> > NAME GETS MISSES IMMEDIATE_GET
> > IMMEDIATE_MISS
> > -------------------- ---------- ----------
> > ------------------------ --------------------------
> > redo allocation 4865380 5915 0
> > 0
> > redo copy 0 0 0
> > 0
> >
> > My current init.ora
> > ========================
> > log_small_entry_max_size 800
> > log_simultaneous_copies 0
> > spin_count 1
> >
> >
>
> It looks like all copies are occurring on your redo allocation latch. The rule
> of thumb is that the misses/gets ratio should be less than 1%, which it is. If
> you wish to do the copies using the copy latch or latches more often, set:
>
> log_small_entry_max_size - (smaller than 800, possibly 400)
> log_simultaneous_copies - (up to 2*CPU count, possibly 4)
>

You might want to play around with SPIN_COUNT as well. I believe the default is 2000 - you might be OK with this with just two processors, but you may get better performance if you bump it up a bit.

SPIN_COUNT refers to how many times a process will "spin" on a particular resource before giving up and blocking against it. Once a process switches over to block, it will check periodically to see if the resource is available. This means that the resource will not be grabbed the very moment that it is available if the resource is being blocked against, whereas it will be grabbed (a spinlock or "latch" will be gotten) the moment it becomes available if the process is spinning on it. If you have a lot of processes blocking on resources, your general performance may degrade. Bumping up SPIN_COUNT would fix this.

Brenda Muller

-------------------==== Posted via Deja News ====-----------------------
      http://www.dejanews.com/     Search, Read, Post to Usenet
Received on Tue Feb 25 1997 - 00:00:00 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US