home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.os.vms
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!sdd.hp.com!ux1.cso.uiuc.edu!news.cso.uiuc.edu!jsue
- From: jsue@ncsa.uiuc.edu (Jeffrey L. Sue)
- Subject: Sometimes documented things don't work (was: Running LOGINOUT)
- References: <9301051432.AA20133@uu3.psi.com>
- Message-ID: <1993Jan5.164927.22609@ncsa.uiuc.edu>
- Originator: jsue@pluto.ncsa.uiuc.edu
- Sender: usenet@news.cso.uiuc.edu (Net Noise owner)
- Organization: The Dow Chemical Company
- Date: Tue, 5 Jan 1993 16:49:27 GMT
- Lines: 76
-
- In article <9301051432.AA20133@uu3.psi.com> leichter@lrw.com (Jerry Leichter) writes:
- >
- >It would be nice if it were decently documented, rather than being hinted at.
- >In general, it's clear in VMS when something is documented/supported, and
- >when it just happens to work. I can think of no other example in which the
- >VMS documentation tip-toes around the edges but is unwilling to really
- >commit itself...
-
- Even stranger, though, are the cases where something is fully documented,
- but still doesn't work - or at least, doesn't work properly. Most notable
- (notorius):
-
- [get ready DEC]
-
- 1. IO$M_FORCEPATH & IO$M_SETPRFPATH -
- This was introduced in VMS 5.4 and never really
- worked correctly for us. Even under 5.5 DEC told us not to
- use it - although they didn't take the articles out of DSN
- that showed example programs of using it.
-
- What happens is that if you set the preferred path in a mixed-
- interconnect cluster, with dual-ported HSC-based disk, the
- satellite nodes will place the disk(s) into mount verification
- if the disk(s) ever fail over to an HSC not designated as the
- preferred path.
-
- To be fair, DEC *finally* did come up with a workaround whereby we
- are to set the preferred path, mount (or failover) the disks, and
- then clear the preferred path. Clearing the preferred path is done
- by setting the preferred path to the current nodename.
-
- 2. SYSMAN STARTUP vs INSTALL.EXE -
- In our VAXcluster we have a large, somewhat complex configuration
- of software, and having a database-driven interface for starting
- them seemed like a great idea. Due to the large number of apps
- the best way to speed up STARTUP was to execute many of them
- concurrently - batch jobs in our case.
-
- Actually, we never actually implemented this under SYSMAN STARTUP,
- but our SYSTARTUP_V5.COM performed many SUBMIT commands in order
- to accomplish the same thing. SYSMAN STARTUP was a future project
- that we intended to pursue to make managing the app startups
- easier (ie., reduce the number of IF...THEN statements necessary
- in SYSTARTUP_V5.COM).
-
- The problem is this: Apparently since VMS 5.1 DEC has known that
- the INSTALL.EXE program would *sometimes* get into a deadlock
- situation if more than one process was trying to add or replace
- an image at the same time. Somehow this never affected us (we've
- been doing submits for a couple of years) until we got faster
- processors (VAX 6000-510 & 6000-610), and upgraded to VMS 5.5.
- It seems to be completely a "timing" issue and I can't really
- explain why it never cropped up before, but it finally got to the
- point where we could *never* run INSTALL from multiple processes.
- This means that many of the benefits of SYSMAN STARTUP are lost
- to us until DEC fixes INSTALL.EXE... not likely to happen in my
- lifetime, so I understand.
-
- NOTE: SYSMAN STARTUP is **fully** documented and we are expected to
- take advantage of it... but it won't help us speed up our boot
- process - and the installation of about 200 software products.
-
- Also, some of our applications startup themselves via detached
- processes, making this kind of difficult to control. What's even
- funnier (NOT!) is that we can no longer use the "documented"
- way to have decwindows startup - ie, it just happens automatically
- from VMS$CONFIG-050_VMS.COM & VMS$LPBEGIN-050_STARTUP.COM. Instead
- we have to DEFINE DECW$IGNORE_DECWINDOWS TRUE, and then SUBMIT
- the DECW$STARTUP.COM file seperately.
-
- The INSTALLE.EXE problem is a known bug, but not a problem, according
- to DEC... GET A CLUE!!!
- --
- -----
- Jeff Sue
- - All opinions are mine - (and you can't have any, nya nya nya)
-