home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!charon.amdahl.com!pacbell.com!decwrl!sun-barr!cs.utexas.edu!zaphod.mps.ohio-state.edu!darwin.sura.net!haven.umd.edu!umd5!bill.ab.umd.edu!bill
- From: bill@bill.ab.umd.edu (Bill Bame)
- Newsgroups: comp.os.vms
- Subject: 2 node LAVC "problem"!
- Keywords: VMS, LAVC
- Message-ID: <17017@umd5.umd.edu>
- Date: 16 Nov 92 22:49:04 GMT
- Sender: news@umd5.umd.edu
- Reply-To: bill@office.ab.umd.edu
- Organization: M.I.E.M.S.S.
- Lines: 64
-
- Ok, I know this has been asked before (probably hundreds of times) here and
- elsewhere, but I never thought I would be in this situation, so I didn't
- bother to read the replys. Before I go into detail, the question is:
- "Is it possible to set up a two node cluster in such a way that crashing
- either node does not reduce the number of votes below QUORUM".
-
- I have a client who has replaced a LAVC made up of 4 overburdened
- microvax IIs with a cluster of two 3100/80s. That's the good news.
-
- +--------+ +-------+ +-------+ +--------+
- |System | |Common | |Common | |System |
- |Disk for+--+Data | |Data +--+Disk for|
- |VAX #1 | |Disk | |Disk | |VAX #2 |
- +----+---+ +-------+ +-------+ +---+----+
- | |
- | SCSI SCSI |
- +-------+ +-------+
- | |
- +---+---+ +---+---+
- |VAX #1| |VAX #2|
- +---+---+ +---+---+
- | Ethernet |
- +----------------+
-
- The 3100/80s have separate system disks (VMS 5.5-1), but need to share queues
- and disks (and tapes if DEC ever decides to support TMSCP-served SCSI tape
- drives). I suggested that they use DFS and DQS instead of VAXCLUSTER since
- there are only two nodes - and since they don't share a system disk. They
- weren't comfortable with that idea, mostly because it would cost them extra
- money, but also because they may want to add more nodes to the cluster in the
- future. I configured the two 3100/80s as a LAVC, and it runs just fine. The
- users seem VERY happy. Now for the bad news...
-
- There is a "perceived" problem caused by having a two node cluster (which I
- warned them about in advance). Because of the way QUORUM is calculated
- [maximum of: (expected_votes+2)/2 and (total_votes+2)/2], the number of votes
- drops below QUORUM if either node goes down. Either node will boot
- independently of the other since the QUORUM will be calculated using the
- (expected_votes+2)/2 formula, but as soon as both nodes are up, the
- (total_votes+2)/2 formula is used. Let's say that EXPECTED_VOTES=1 and
- VOTES=1 (on both nodes). That means that QUORUM is 1 [(1+2)/2] for a single
- node, but changes to 2 [(1+1+2)/2] once the second node has booted. Simple
- algebra shows that the (total_votes+2)/2 formula won't allow a node to drop
- out of the cluster without "blocking activity" once both nodes have been booted.
-
- Is there some trick (with a quorum disk perhaps) that will make this behavior
- go away?
-
- I realize that I should know this already (having set up dozens of clusters),
- but I have never set up one with less than 3 nodes, and the manuals haven't
- been much help.
-
- Thanks!
-
- -Bill
- --
- --------------------------------------------------------------------------------
- William P. Bame | sneakernet: William P. Bame
- internet: bill@office.ab.umd.edu | M.I.E.M.S.S.
- bbame@achi1.ab.umd.edu | 22 South Greene Street
- AT&T : [Work] (410) 328-3062 | Baltimore, MD 21201
- CIS : 71620,425 | Room: T1R55
- --------------------------------------------------------------------------------
-
-