From: young_r@eisner.decus.org
Sent: Tuesday, July 18, 2000 12:22 PM
To: Info-VAX@Mvb.Saic.Com
Subject: Re: Looking for information on PE parameter usage

In article <00256920.0051FA42.00@quegw01.btyp>, Steve.Spires@yellowpages.co.uk writes:
> cc:
> bcc:
> Contact:   Tel: 3063  -  VSSG, 1st Floor, Bridge Street Plaza
> 
> Looking for information on PE parameter usage
> 
> Can anyone point me in the right direction for help on the PE parameter settings
> in SYSGEN? I see that they are reserved parameters, but I wonder of some could
> help us with Oracle/VMS/Cluster prefromance problems.
> 

	You know.. Deja is still down and it stinks up the house.

	Fortunately, I occasioally save a gem.  Keith posted this
	out here March 1999.  This has been discussed on occasion
	and George Cornelius points out he sets PE1 to 1000 (I believe).

	You may wish to contact Keith directly.  He may have a tool
	to automate the checking by now :-).  If so, let us know
	what you find.

				Rob

---

PE1 controls the maximum size of a lock tree that is allowed to be
remastered _off of_ the node on which it is set.  (It has no effect on
the movement of lock trees _onto_ the node.)  It is most often needed in
clusters where 2 or more nodes of equal horsepower run the same workload, and
lock mastership of critical lock trees thrashes back and forth between nodes.
If the lock trees are fairly large, it can take a fair amount of time to
remaster them, because the algorithm (at least as of 7.1) moves one lock per
SCS message. (I've observed I/O pauses of 10 to 50 seconds caused by
unthrottled lock mastering in a cluster with several lock trees of 15K locks
each.)

A fairly simple way to see if and when you're getting remastering activity is
to check the counters which track remastering messages:

$ analyze/system
eval @pms$gl_rm_rbld_rcvd  !Count of remastering messages received
eval @pms$gl_rm_rbld_sent  !Count of remastering messages sent

By doing this on each node, and watching the values over time, you can
see where lock trees are being moved to and from.  If you see a pause
in I/O, you can check the counters to see if they've recently incremented
by a large amount, and thus tell whethern or not remastering caused the pause.

PE1 is a dynamic parameter, so you can set it to a small non-zero value
and see if that helps, then set it back to zero if it doesn't.
-----------------------------------------------------------------------
Keith Parris|Integrity Computing,Inc.|parris@decuserve.decus.org-nospam
VMS Consulting: Clusters, Perf., Alpha porting, Storage&I/O, Internals

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own