home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.os.vms:17697 vmsnet.mail.misc:92
- Path: sparky!uunet!verifone!jimmy_t
- From: jimmy_t@verifone.com
- Newsgroups: comp.os.vms,vmsnet.mail.misc
- Subject: Re: How to make VMSMail run lots faster
- Message-ID: <1992Nov9.160445.4632@verifone.com>
- Date: 9 Nov 92 16:04:45 -1000
- References: <1992Nov6.190229.4628@verifone.com>
- Organization: VeriFone Inc., Honolulu HI
- Lines: 45
-
- Nick Abbot asked me to post the folowing:
-
- In article <1992Nov6.190229.4628@verifone.com>, jimmy_t@verifone.com writes:
- > We are having severe VMS Mail performance problems. The problems center around
- > users with lots (thousands) of mail messages stored as external files in the
- > VMS directory where they receive mail. Once the size of this (or any) VMS
- > directory passes 128 blocks or so (many of ours are >> 1000 blocks),
- > performance for adding new files begins a rapid decline. Other users sending
- > mail to these accounts will encounter seemingly random long delays waiting for
- > the "MAIL>" prompt after sending.
- >
-
- "Making VAX Mail Fly". Would that be, "VAX Air-Mail"...Hmmm?
-
- It seems to me, that you have a similar problem that Pathworks for VMS has.
- Lots of files. One of our clients bought a 3100-90 for a file server. The
- problem is the PCFS_SERVER's inability to cache directory info. I suspect this
- is where your DIO's are being spent. I've implemented a MASSIVE increase in
- the VMS ACP sysgen params that cache this on all drives mounted/system, it's a
- shared cache. ACP_DINX*, ACP_DIR* and ACP_HDR*, the results were staggering,
- NO delays with the C:\> DIR /S. Some Wordperfect(dos) functions that do
- directory searches, hauled donkey. MONITOR FILE...was pegged at 100% hit rate.
- Really, really zippy.
-
- Keep in mind these params come out of paged pool, so for every page of
- increase, bump up PAGEDYN by number of bytes. Also bump up the system working
- set by the same number of pages...SYSMWCNT. That will prevent system page
- faults. I'll be the first to admit that not all performance issues can be
- solved by throwing memory at them, but this one was obvious to the most casual
- observer, and I had a few lingering about at the time.
-
- In the case of the file server. I had to reduce the memory for the
- PCFS_SERVER process to make room, I originally gave it the WORLD. The end
- result was well worth it. Usually there was a pregnant pause while doing a
- directory of a file service with hundreds of files in it. That pause went
- away.
-
- On a side note. You might want to BAC/IM these drives, and INIT them
- with the /HEADERS=<very big>, ie. double the size of INDEXF.SYS now. This will
- give your INDEXF.SYS the room to grow and not "fall apart" later. Then BAC/IM
- to restore, using the /NOINIT qualifier to preserve the newly allocated size of
- INDEXF.SYS. DIR/SIZ=ALL.
-
- Nick Abbot 70303.1354@compuserve.com
-
-