INTNER@CUTCV1.BITNET (05/13/86)
After sending out 6 or 8 individual copies and losing some one's
address, i decided to send the whole mess to the net.
/jsi
My feeling from the messages I received is the numbers I got (3 hours
for a Standalone Backup on a 750 using the TU58) is pretty much in the
ballpark. There also seemed to be a lot of things I could do to
improve it.
1) The first would be to do a full backup and restore to
remove any fragmentation of the disk.
2) Next would be to specify a Buffer_Count of 5 (something
like blocking factor, i guess).
3) Another good suggestion was to build the Standalone Backup
kit on the RA81, instead of the TU58.
4) Good things to do would also be to get a removable disk,
and do the backup on that, or get a 6250 tape drive.
The main suggestion that i think i will follow is to do Standalone
Backups only when i upgrade VMS versions or install new software. i'm
going to have to build some procedure (it should be very
straightforward) to do a backup of the user area of the disk only,
since that is the only area that changes on a regular basis (or should
be anyway).
below i will include all of the responses i received (for your perusal
and gratification).
jonathan intner
computer system manager, CCIMS
BITNET: INTNER@CUTCV1
Internet: G.Intner@cs.columbia.edu
---------------------------------------------------------------------------
Subject: Standalone Backup question
For some reason, during a recent standalone backup, I had to specify
/density=1600 in the command line. I had never had to do this before
and I can not remember why this was required all of a sudden. The
problem was that the tape drive kept defaulting to 800 BPI (TU77)
where it used to always default to 1600 BPI. I at first did not
realize that the drive was at 800 and the backup was taking much
(twice) longer that usual.
We do stand alone of both our drives (RM80 and RA81) on our 780
once a month.
---------------------------------------------------------------------------
We are a new VAX site. We have two DEC-20s that together serve
about 120 simultaneous users (and perhaps 500 altogether). We
acquired a VAX 8600 last fall and it now has three RA81s and one
RA60 (on which we mount two different packs).
Right now, only our system disk and one of our RA60 packs are used
daily (most of our users are still on the DEC-20s).
Our backup schedule is as follows:
weekly
full save of all disks
daily
incremental save of the system disk and the
active RA60 disk
We do not do the full saves using standalone backup. Instead,
we use
BACKUP/IMAGE/IGNORE=INTERLOCK ...
I know of several sites who also follow this procedure. It is
too inconvenient to end timesharing just to perform a system save.
I do take a standalone backup before installing new versions of
VMS itself or when compacting the system disk (which we did once
as an experiment).
---------------------------------------------------------------------------
Subject: Standalone backup question
It takes us about an hour and a half on a 785 with a ta (6250) drive
to do an image backup on about 400K. To the best of my recolection,
it took 6 tapes and 5 hours when we had to tu drive.
We are now clustered, so we only do a standalone or vms backups when
we want to get rid of fragmentation of a disk. For our regular backups,
we do an hsc backup .. takes about 5 minutes!
-------------------------------------------------------------------------------
>I just did a Standalone Backup of our only disk (an RA81) on our 750.
>It took (in my mind) an unusually long amount of time. It took 3 and
>a bit tapes, and a total of 3 hours (clock time). The tape drive is a
>TU80 (i think -- it is 1600 bpi, i'm sure). The RA81 has a total of
>300,000 blocks used.
>
>Is it me, or do other people's out there backups take that long?
>Anyone have any suggestions?
>
We have a similar configuration and in fact the same problem. probhably
you were using the verify option, didn't you? If not, you should do so.
I could reduce the amount of time by giving the maximum valid buffercount of 5.
My command looks like this:
$BACKUP/IMAGE/VERIFY DUA0: MSA0:GAMMA_270486/BUFFER_COUNT=5/LABEL=BT0011/REWIND
This took me from 36 to 42 minutes per tape with the same amount of tapes.
>While I have your attention, how do most people do backups (like how
>often? Standalone? only the UserDisk? only after software
>installations, etc), or does it depend to much on the specific needs
>of your site?
>
On our two 780's with two RM05 each, we are making an image backup to disk
every week with a turnaround of three weeks.
An incremental backup to tape is done every morning.
On the 750, where backup's take too long, we only backup the whole system
before (not after) software installations. Backup there is done on a per
project basis. We only back up the user's important data. The project's
leader has to tell us which directories and/or files, how often and on how
many tapes he wants to backup. This information is written into a file. A
selfmade utility (command procedure) is daily invoked by the operator, which
then performs the backup to the project's set of tapes. Currently, five to
seven projects (with a total of 50 users) share that machine.
>if there is any interest, i will be happy to summarize for the net.
>
Please, send me a summary, because backup is always to take serious.
-------------------------------------------------------------------------------
3 Hours for 3 reels are really too much. I have also a 750 with Tu80,
and it takes not longer than 20 minutes per reel. When you do your next
Stand-alone, please take a look on the tape and on the disk READY lamp. If
the tape starts and stops quite often, and the READY bulb flashes all the time,
then your disk files are probably very fragmented. If this is the problems,
then the cure is to backup the whole disk, using STAND-ALONE only, and
immediately restoring it. This will create the new files contigous. Don't
forget to use the /IMAGE qualifier, and when restoring - /RECORD.
For the frequency of Stand-alones - It depends on your site's
needs. I am doing it every one month (incrementals every 2 days), and
most important - Before any software installations. It is good also
to back-up after the installation, but it is more important to do it
before, in case the installation fails.
-------------------------------------------------------------------------------
If you have a lot of space in large files, you can improve the
performance of a streaming tape drive like a TU80 by increasing
Backup's buffering. Add /BUFFER_COUNT=5 to your command string.
Unfortunately, Backup allows a maximum of only 5 buffers.
The default is 3.
On the other hand, if most of your files are small, Backup will be
spending most of its time waiting for the disk heads to reposition.
This causes problems for a streaming drive, since the data will usually
not be ready in time for the drive to stay in streaming mode.
Adding the /FAST qualifier supposedly helps by telling Backup to work
directly from the index file rather than from the directories.
I think it does help a little.
The best we have seen for BACKUP/IMAGE is about 1/2 hour per tape
on a TA78 at 6250 bpi, which uses 4 tapes for a 90% full RA81.
We maintain Backups of 3 RA81s (1 system, 2 user). We have other
disks, but we don't maintain Backups of them. Writing 12 tapes
takes at least 6 hours, and usually more, since we can't afford to
take our 8600 standalone. It accumulates only about 1 hour of
idle time per week.
BACKUP/PHYSICAL only takes about 15 to 20 minutes per tape, but we have
essentially no use for it. Our users need to be able to restore files
on an individual basis. Accidental deletions happen much more often
than the complete loss of a disk drive. The only time I have used it
was when DEC was upgrading our RA81s from rev5 to rev7. To minimize
system down time, each drive was copied to one just upgraded. A full
drive-to-drive copy only took about 20 minutes. (The disk controller
is an HSC-50.)
As a result, we do full Backups (/RECORD) once a week, and incremental
Backups (/SINCE:RECORD) on a nightly basis. By the end of the week,
our incremental Backups occupy about 1.5 tapes. Each daily Backup
includes all the files created since the weekly Backup.
The Backup of the system disk is done separately from the user disk
Backup, as VMS Backup doesn't allow you to specify the name of the
source disk when restoring from tape. The user disks are a bound
volume set, so there is no need to specify which of the two a file
was on originally.
I hope this is some help.
-------------------------------------------------------------------------------
We have a backup system that I am rather proud of -- it works smoothly and
with little manual intervention. We don't do standalone backups, except
after upgrading VMS, of course. We have an incremental backup batch job that
runs every night at 4 a.m.; it purges all files to 3 versions, selected file
types (EXE,OBJ,LIS,etc.) to 1 version, and does an incremental backup to
RL02. The RL02, of course, only holds 10 megabytes, but even so a single
disk usually lasts a week or two. We have 6 RL02 cartridges, used in rotation.
We also do full backups to tape, currently once a month the day before PM.
This takes several hours and 4 tapes at 3200 bpi (wierd double density
Cypher drive, great for backup, also does 1600 bpi of course). When we
do such a full backup, we then cycle to the next incremental backup
cartridge.
Net effect is that we have backup every day and can lose no more than a
day's work, even if we lost all of both disks (we have 1 130 Mbyte drive
and 1 500 Mbyte drive, each is "mapped" to appear to be 2 drives). It strikes
me that a bit more explanation is in order. Our logical disks:
DRA0 65 Mbytes System disk, routinely backed up
DRA1 65 Mbytes Swapping disk, not backed up
DRB0 250 Mbytes User disk, routinely backed up
DRB4 250 Mbytes Applications disk, backed up as needed
DRB4 contains several applications directory trees -- TEX, VLSI design
tools, Template graphic system, etc. -- which are too large to fit on
the system disk but do not change very often. These we back up on a
case by case basis, whenever the tree is updated. It also contains
scratch. Earlier comments about routine backup apply to DRA0 and DRB0.
We also cycle full backup tapes among 3 sets, and every once in a while
we do an "archival" backup, which is merely a full backup that we don't
cycle but keep in a separate building for at least a year, in case of
earthquake or other disaster.
Hope this helps.
------------------------------------------------------------------------------
BACKUP efficiency is sensitive to a number of factors. Tape block
size is one of the more important ones; the following suggests the
largest (and most efficient) blocks for each density:
800 bpi 8192 ( 8Kb) bytes/block (default)
1600 bpi 16384 (16Kb) bytes/block
6250 bpi 65024 (63.5Kb) bytes/block.
The larger the tape blocks, the fewer I/Os are required, and the
faster things will happen. I believe TU80s are streaming drives, so
it may also help to specify the maximum 5 buffers. (Note that 5*63.5K
equals 635 memory pages for buffers alone, so allow a large working
set.)
/IMAGE should be used for full backups, and usually /FAST for
incrementals. (Standalone BACKUP uses /IMAGE by default.)
We have Flip and Flop Fulls every weekend, and Flip and Flop
Incrementals every evening. These are done online for all but scratch
disks.
-------------------------------------------------------------------------------
>I just did a Standalone Backup of our only disk (an RA81) on our 750.
>It took (in my mind) an unusually long amount of time. It took 3 and
>a bit tapes, and a total of 3 hours (clock time). The tape drive is a
>TU80 (i think -- it is 1600 bpi, i'm sure). The RA81 has a total of
>300,000 blocks used........
Three hours isn't too bad. Our RA81, with about the same disk usage, takes
about 5 hours and, at 1600 bpi, as many as 12 tapes!!!! I wouldn't worry.
We do daily incrementals four days a week, Mon - Thurs, and a weekly
incremental on Friday. Every four weeks, if we dont run into production
dates, we do an image. This schedule is just for our RA81s. We use
the same for other disks on the system but stagger them.
------------------------------------------------------------------------------
Your configuration sounds a bit like mine (one RA81, TU80) so let me
offer my suggestions.
First, avoid loading Standalone Backup from TU58. To do this you can
build a Standalone Backup system on your disk with
@SYS$UPDATE:STABACKIT. When asked for the name of the device on which
to build the kit respond DUA0:. This will create a Standalone backup
system in directory [SYSE]. This system can be booted with the command
B/E0000000. Much faster than TU58.
Otherwise, assuming you loaded Standalone Backup from TU58, 3 hours
start to finish sounds about right. As an estimate I usually figure
about 25 minutes per tape, but this can vary depending on what your
files are like. Large files go very quickly, streaming the TU80 almost
constantly, but bunches of small files take much longer.
Also, insure that files like PAGEFILE.SYS, SWAPFILE.SYS, and SYSDUMP.DMP
are marked NOBACKUP (SET FILE/NOBACKUP filename.ext) so the data is not
written to the saveset, only the header info. VMS 4.1 seems to set
these files nobackup automatically, but you must do this yourself under
3.x.
If you have the luxury of two tape drives you can use both (writing one
while the other unloads) by specifing the destination as
MSA0:savesetname.bck,MSB0:
Frequency of backup, as you say, depends much on the needs of your site
(how critical is the data, how often does it change). I use a system of
infrequent full (image) backups of the disk (one disk, both system and
user data) and daily incremental backups (DUA0:[*...]/SINCE=BACKUP) on
the theory that I am protecting against not only total hardware
failures, but software and user failures (files become corrupted, a user
accidently deletes the wrong file etc...). Using this method I can
recover data to a specific point in time. The cost of course is more
tape, and should I ever have to recover the whole disk it will be a much
longer process because of the many incrementals to be restored.
If you do get sufficient response to your question I would be interested
in seeing the summary posted
------------------------------------------------------------------------------
At the place where I work (not OSU), we have three 750s, two uvax-Is,
and a 730.
All of the machines except for one 750 are test machines (i.e., the
stuff on the disk doesn't change very often). These are backed up
fully (image) whenever the OS or layered products change (updates/
installs).
The main 750 is the center of our Software Development/Office
Automation. It has an RA-81 (combination system/data disk), an
eagle (data disk) and two (logical) smaller SI disks (one a paging/
swapping disk, the other a scratch disk) [there're also some smaller
disks, but they don't count :-]. The RA-81 and eagle have
an image backup performed weekly, with incrementals on a daily
basis. (note that this is done with multi-user VMS up and
running)
Every once in a while the RA-81 will get extremely fragmented
(as determined by INDEX or PIP). At this point, I get out the
old standalone kit and compress it. This usually takes about 14
hours (RA-81 about 7/8 used. TU-78 @ 6250 BPI. Backup with
verification).
I'd be very interested in what other sites do.
------------------------------------------------------------------------------
Hi,
On one of our 750 systems, with a tu80 at 1600 bpi, the backups
take about 4 hours or so. Of course, that disk is almost full, so we're
writing approx. 700,000 blocks or so.
About when we do backups... Usually there is an incremental done
every day, then at the end of the week, usually Friday, a backup is done
to get everything since the previous Friday. Full image backups are done
once a month, usually the last weekend of the month. Doing it this way,
with the Special Friday, when we do restores, we just have to take the
full backup and then the previous friday backup.
When we do the backups, everything is backed up, from the system
files to the user files.
-------------------------------------------------------------------------------
I would say that what you experiance is an accurate account
of backing up your system
We have a 730 with ra81 and find our backups take just a bit
longer
As to the neccesity of always doing standalone backups -
I feel it is only needed after an update to vms.
Otherwise, on on-line image_backup is all that would be
needed to restore the entire system.
backup/image dua0: msa0:full.bck/save/dens=1600
this minimal command (set login/inter=0 before, and try
to minimuize open files!!) will work always...