[ From:	CRDGW2::CRDGW2::MRGATE::"SMTP::CRVAX.SRI.COM::RELAY-INFO-VAX"  8-FEB-1991 23:27:00.19  To:	MRGATE::"ARISIA::EVERHART" CC:	) Subj:	Re: Which node do I bring up first?   * Received:  by crdgw1.ge.com (5.57/GE 1.80)( 	 id AA04938; Fri, 8 Feb 91 22:14:09 ESTY Received: From UCBVAX.BERKELEY.EDU by CRVAX.SRI.COM with TCP; Fri,  8 FEB 91 19:03:02 PST , Received: by ucbvax.Berkeley.EDU (5.63/1.42)) 	id AA11763; Fri, 8 Feb 91 18:58:37 -0800 9 Received: from USENET by ucbvax.Berkeley.EDU with netnews . 	for info-vax@kl.sri.com (info-vax@kl.sri.com); 	(contact usenet@ucbvax.Berkeley.EDU if you have questions)  Date: 6 Feb 91 23:36:43 GMT t From: zaphod.mps.ohio-state.edu!unix.cis.pitt.edu!pitt!cuphub!edinboro!gcc!brown@tut.cis.ohio-state.edu  (The Raven)0 Organization: Grove City College, Grove City, PA, Subject: Re: Which node do I bring up first? Message-Id: <374@gcc.uucp>2 References: <1991Feb1.135720.13902@axion.bt.co.uk># Sender: info-vax-request@kl.sri.com  To: info-vax@kl.sri.com   2 In article <1991Feb1.135720.13902@axion.bt.co.uk>,* pkatz@axion.bt.co.uk (Philip Katz) writes:  A > 	Our main cluster will soon comprise a 6310, a 6410 and a 6510.  > F > 	I have heard suggestions that when booting the whole cluster, it isJ > advantageous to boot the most powerful processor first, as then the lockG > manager will run on it and this is a good and useful thing to happen.  >  > 	Is there any truth in this? >   O Yes, and no.  It was much more significant prior to VMS 5.0.  Nowadays the lock O load is much better balanced dynamically between all nodes.  If you do run into O problems, you can tailor it using the LOCKDIRWT sysgen parameter.  For example, N if you want the load to be primarily handled evenly by your 6310 and 6410,withM a little bit of load handled by the 6510, you can set the LOCKDIRWT parameter N to 2,2 and 1 respectively.  However, a warning:  The default for the LOCKDIRWTL parameter is 0, so if you change it on any node, you had better change it onN another - 2,2, and 1 will work, but 1,1, and 0 will mean your 6510 will handleH almost no locks - which can cause some nasty hangups if the 6510 is mscpM serving local disks around the cluster.  It won't really matter if you aren't F MSCP serving disks in that manner.  I have a rule of thumb that I use:F 	1.  If the node MSCP serves disks to other nodes, LOCKDIRWT should be?             equal to the highest weight used on any other node. F 	2.  If the node is only an MSCP recipient node and does not have realG 	    time programs that do a lot of locking (such as a VS3100 in a LAVC D 	    configuration), set lockdirwt to 0 for that node and 1 for all G 	    others (unless circumstances show a need for different balancing). O This is particularly significant in LAVC systems that have VS3100 type boxes as N cluster members.  If you have, for example, 2 6410s and a VS3100 in a cluster,P with LOCKDIRWT = 0 on all three nodes, MSCP served disks on the VS3100 can have K locks mastered on the 3100.  As long as the 3100 is the only one using that O file or record, no problem.  But let's say that you have a big database file on I your 6410 that is accessed via MSCP serving on the 3100.  The 3100 runs a K process that is the first one to open the database.  The master copy of the M lock for that file is then held on the 3100, and any other operations on that O file will have their locks mastered on the 3100 including ANY AND ALL PROCESSES J ON THE 6410s that access the file!  This means that for any operation thatL requires lock conversion or queueing, LAVC activity will occur for all theseO processes.  This can be a major problem.  However, if you set LOCKDIRWT on your K 6410s to 1, and to 0 on the 3100, the 3100 will only master locks for disks  that are local to it.   J I hope this helps - if you need any more info on it, locking is covered inL pretty good detail in the VMS internals and Data structures reference manual Version 5.0.   --  L     +----------------------------------------------------------------------+L     |   Mitchell W. Brown              Internet: brown%gcc@edinboro.edu    |L     | Grove City College TLC           Uucp: ...pitt!edinboro!gcc!brown    |L     |  Grove City, PA 16127            (412) 458-2072                      |L     +----------------------------------------------------------------------+  