Re: dump scheduler algorithmn

2000-12-20 Thread Chris Karakas

Amanda Backup wrote:
 
 My configuration (2.4.2) has these settings:
 
 dumpcycle   1 weeks
 runspercycle6
 tapecycle   18 tapes
 
 Does this mean that given a level 0 today, the
 next level 0 will be at the latest 1 week from
 today?  

Exactly. Of course, if the balance of load over the week (i.e. over the
dumpcycle) permits it, you might see level 0s more often.

 Or could it mean that within each dump cycle
 there will be at least 1 level 0 but that
 they could be up to 12 dumps apart (begin
 of one cycle, end of next)?  That is nearly
 what one file system is doing.
 

Normally, this shouldn't happen. That it happens, it means that the load
is too heavy for this dumpcycle. Now, what does this mean, you will
ask... ;-) It means that the dumpcycle is too short, or the tapes have
too small a capacity and they cannot absorb more data per run. The
easiest solution is to increase dumpcycle to, say, 10 (you can verify
that it is indeed a balancing problem if you do a "amadmin config
balance" and see the projected load and the number of overdue disks).

What AMANDA always tries is to get the load as evenly distributed as
possible over the dumpcycle period, taking into account constraints like
"at least one level 0 every dumpcycle days per disk", priorities,
capacities, previous backup sizes  etc. This is a "bin packing problem",
the "bins" being the tapes :-). There is no exact solution to it -
actually, it is a computationally very hard one and the subject of
ongoing academic research. But if you choose your parameters
"reasonably", you will see that AMANDA solves it very well (I am always
amazed by the combinations she uses to "pack" all those levels in one
tape efficiently). 

I suggest you start with a somewhat larger dumpcycle than you think is
necessary, say 15 days. Let AMANDA do some dumpcycles this way. You will
get a feeling of the "balanced load" on your tapes, i.e. you will see
that your tapes get *evenly* filled (as long as this is possible of
course), up to a certain percentage. AMANDA will *not* try to fill the
tapes as much as possible because she tries to spread the backups evenly
on the tapes. As soon as you see that you have steadily "room" on your
tapes, that is left empty, you can start decreasing your dumpcycle
carefully. This way you will eventually reach the Optimum :-)

-- 
Regards

Chris Karakas
Dont waste your cpu time - crack rc5: http://www.distributed.net



dump scheduler algorithmn

2000-12-19 Thread Amanda Backup

Given my results with one particular file system I
don't think I understand dump/run/tape cycle.

My configuration (2.4.2) has these settings:

dumpcycle   1 weeks
runspercycle6  
tapecycle   18 tapes

Does this mean that given a level 0 today, the
next level 0 will be at the latest 1 week from
today?  For most of my file systems that is
what I'm seeing.

Or could it mean that within each dump cycle
there will be at least 1 level 0 but that
they could be up to 12 dumps apart (begin
of one cycle, end of next)?  That is nearly
what one file system is doing.

Here is the amoverview for the strange fs.



   date   11 11 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12
host   disk   29 30 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19

butch  /w  0  1  1  1 1  2  2  2  2  0 1  1  1  1  1  1 1  0

   -- 10 days, 9 dumps apart -- -- 10 days, 8 dumps apart --


Is this normal, nothing to be concerned about?

For the other 7 file systems, the intervals between
level 0 dumps were only 2 to 5 dumps (the latter only twice).

All from the same host, the tapehost, so availability was
not a problem.  Plenty of holding space and tape capacity.

Comments?

jl
-- 
Jon H. LaBadie  [EMAIL PROTECTED]
 JG Computing   [EMAIL PROTECTED]
 4455 Province Line Road(609) 252-0159
 Princeton, NJ  08540-4322  (609) 683-7220 (fax)