NetNews Usenet Archive 1992 #27/NN_1992_27.iso/spool/comp/os/vms/18251

Internet Message Format | 1992-11-20 | 3.6 KB

Path: sparky!uunet!spool.mu.edu!sdd.hp.com!news.cs.indiana.edu!nstn.ns.ca!ac.dal.ca!ae
From: [e-mail]
Newsgroups: comp.os.vms
Subject: BACKUP performance - can I expect any better?
Message-ID: <[e-mail]>
Date: 18 Nov 92 19:58:35 -0400
Organization: Dalhousie University, Halifax, Nova Scotia, Canada
Lines: 56

  After I'd prepared this note one of my colleagues who is more up to date on
his comp.os.vms reading pointed out that a similar question on BACKUP
performance was asked and had been given some responses just a few days ago.
I hope readers won't mind my going over the ground again in some more detail
in case I can elicit any new insights.

  I've been doing some rough calculations on the performance of our backups.
These are done on a VAX 4500 onto a TSZ07 nine-track tape drive. The BACKUP
command looks like

    BACKUP /IGNORE=INTERLOCK disk tape_file /BLOCK_SIZE=32256

WSMAX on this system is 65500.  The quotas for the username that runs BACKUP
are

    Maxjobs:        20  Fillm:       128  Bytlm:        65536
    Maxacctjobs:     0  Shrfillm:      0  Pbytlm:           0
    Maxdetach:       4  BIOlm:       128  JTquota:       1024
    Prclm:           5  DIOlm:      4096  WSdef:          200
    Prio:            4  ASTlm:      4096  WSquo:        16384
    Queprio:         0  TQElm:        10  WSextent:     24000
    CPU:        (none)  Enqlm:       256  Pgflquo:      32768

  A recent backup took four hours and nine tapes to back up 2.5 million blocks
off an RA92 connected via a KDA50 controller.  This equates to a data rate of
about 89,000 bytes per second ((2,500,000 / (4*3600)) * 512).  This includes
time taken for rewinding and mounting tapes.  Given that the TSZ07 is rated at
200 inches per second rewind time and, by observation, takes 55 seconds to
load a tape there is a minimum overhead of 200 seconds per tape.  Say we allow
300 seconds per tape to account for operators not being right on top of
things, the data rate goes up to about 105,000 bytes per second.
  Another backup of 3.9 million blocks off an RF73 connected via DSSI took 3.5
hours and 13 tapes for a data rate of about 159,000 bytes per second.  Using
the same overhead as with the previous example this goes up to about 230,000
bytes per second.
  Although the two backups were done under somewhat different conditions, the
RA92 on a Tuesday evening and the RF73 on a Saturday morning, the data rates
from these two examples don't seem inconsistent with each other because, again
using rough measurements, the DSSI connected RF73's seem to handle about twice
the I/O rate of the RA92 KDA50 combination.  I've used MONITOR DISK and
MONITOR DISK /ITEM=QUEUE and the average queue on RF73's seems to be about
half the RA92 queue for the same operation rate.
  I'm wondering if a data rate of 100,000 to 200,000 bytes a second is pretty
much the best I can expect.  We're considering going to 4mm, 8mm, or TF85 for
a backup device, so we're looking at rated maximum tape data speeds of about
180KB, 240KB, or 800KB a second.  If we are not getting near the TSZ07's rated
speed of 625KB a second, does it make a difference what kind of tape device we
use.  One big advantage, of course, of a cartridge tape over nine-track is the
great reduction in operator intervention.
  These backups do have to run while significant numbers of users are logged
in.  The batch job that does the backup has the same base CPU priority as the
interactive users. The disks are not badly fragmented.  The RA92 in the
example had about 84,000 files, the RF73 about 100,000.

Aidan Evans ([e-mail]), Computer Facilities & Operations, University
Computing & Information Services, Dalhousie University, Halifax, N.S., Canada