Path: utzoo!mnetor!uunet!husc6!mailrus!tut.cis.ohio-state.edu!bloom-beacon!gatech!mcnc!decvax!ucbvax!uhvax1.uh.EDU!COSC1Q0
From: COSC1Q0@uhvax1.uh.EDU ("Lee K. Gleason, Control-G Consultants")
Newsgroups: comp.os.vms
Subject: VAX Cluster tuning
Message-ID: <8804101421.AA09476@ucbvax.Berkeley.EDU>
Date: 8 Apr 88 18:29:00 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 30

>>We  recently added a VAX 8650 to our cluster of 2 11/780's to be used as a
>>batch engine.  The 780's were operating with about 2% null time, and now are
>>about 70% null, but the response time seems to be slower.  Interactive use
>>hasn't increased, and there are no batch jobs on the 780's.  Am I dreaming?
>>
>We have observed a 785 get very slow after it was booted.  This machine is the
>boot node for a LAVC.  When the 785 is booted, one of the microVAXen takes
>over the handling of the cluster-wide lock management.  Rebooting all of the
>slower machines made the 785 faster, since it took over lock management in
>the absence of any other systems.
     
>We have been warned about this possibility on our CI cluster after we install
>database software on more than one node.


  The sysgen parameter LOCKDIRWT can be used to control which nodes are
"mastering" the locks in a cluster. Higher values give a higher
"mastering" priority. THis would allow you to attempt to get all of the
locking concentrated back on the 785, if you want it there.
  What I can't remeber, is at which version of VMS this feature is implemented. 



Lee K. Gleason
Control-G Consultants.
2416 Branard #D
Houston TX 77098
Phone 713/528-1859

	"One step done, and another begun, and I wonder, how many miles?"