Ok Jon;
>Clarification please; do you mean you have 15 external hard disks, each
>matching the specs you listed yesterday, i.e. 400GB, 6x56GB virtual tapes?
>(btw 6x56 = 330 - what about the other 70GB?)
Well, actually 56 * 1024 - 57344 * 6 = 344064. When FreeBSD (6.1) formats a
400G hd, it yields 348730 of usable area. In order not fill it to the riim,
I took 4G away from the tape setup.
So not 70G but 51270M is taken. I figure FreeBSD needs that for superblock
backup of such a big disk.
>That would be an awesome setup.
It is really. The machine is an AMD84 dual core with 2G ram. 2 IDE drives.
One 80G for the OS and another 120G just for /amanda/tmp (debug files);
/amanda/gnutar-lists (the backup machine mounts an smb share from a windows
machine and backs it up as a client also!) ; /amanda/hold (obvious);
/amanda/split (case I need it, never used it though) and finally,
/amanda/daily, the config folder. And of course, a 400G SATA drive, mounted
as /bck/daily, where tape1-6 are (slots 1-6).
>Do you rotate them to online status such that only 6 vtapes are available
at one time? Why?
Yes !
>Never more than one hard disk online at a time? Why?
Because the possiblity of lossing the data is reduced to almost nothing if
each disk(vtapes) remains on the machine for the smallest time possible.
Each disk is labled outside and inside./bck keeps files (created by
hd_switch and the other utilities) that hold the history of that drive until
it is recycled again.
After the tapes are full, that drives goes to a safe and the next comes in.
So, 1 hd = 1 dump cycle.
>You recreate the virtual tapes each time you rotate a hard disk in? Same
labels each time? Why?
Yes, same labels everytime. Because each hd is numbered and I have control
of the dates each hd was inserted into the system, how many restores were
done with it and how many dumps were ran in it, full or incremental.
>> hd_restore: prepares the system for a restore session with the HD just
>> inserted, preserving the current HD daily folder
> What is a HD daily folder?
Sorry. I tried to keep the explanations short.
When hd_switch prepares a drive, it creates a fresh folder called
/amanda/daily, where all the info and config related to that drive is keept.
If hd_switch detects that the HD was used before, it grabs its label and
re-formats it. If not (new and unformatted) it asks you for a LABEL that is
not registered on the system.
But before doing that, it moves the current daily folder (related to the
currently inserted HD) to folder called
/amanda/daily.DIARIO-01.20060601.16:01, where all indexes, currinfo and logs
will be preserved. Let me explain the folder name:
daily. - config folder
DIARIO-01. - disk label
20060601.16:01 - date & time of insertion.
This folder will be kept until this hd is recycled again (re-formatted).
the hd_backup script, just before it starts the dump, copies /amanda/daily
to daily.current, because
/amanda is also backed up (/amanda/hold, /amanda/daily excluded).
>That will be a long list a couple of years from now :)
No. Because of the reasons above, I will only have 15
/amanda/daily.DIARIO-0x.xxxxxxxx.yy:yy folders always. A re-inserted hd will
loose its data so there's no reason to kepp the amanda records that refer to
it.
>> hd_return: returns the system to the state it was after a restore.
>I suspect you mean "after a restore session started with hd_restore,
>return the system to the state it was before the restore session".
Any switching of Hds requires the machine to be off (no hotswap :( ). THe
hd_ scrpts assume that you know what you're doing.
When you run hd_restore, it assumes that you turned off the machine and
inserted the HD with the data to restore. It also moves /amanda/daily and
/amanda/tmp to /amanda/daily.current and /amanda/tmp.current, and grabs the
/amanda/daily.DIARIO-0x.xxxxxxxx.yy:yy relative to the drive you just
inserted and turns it into /amanda/daily, as if that hd WAS the current HD.
Do whatever restores you need with it.
Then, turn off the machine, get the current HD back in and run hd_return. It
simply moves /amanda/daily.current and /amanda/tmp.current back to
/amanda/daily and /amanda/tmp. But not before checking if the HD you just
inserted back is truly the current HD that was there before.
>It seems to me you could do something similar and reduce or eliminate
>the need for the maintenance scripts.
You're probably right, but with that much HD rotating, and after
experimenting A LOT, the scripts really made the job a lot easier, not only
on the movement of each HD but also on retrieving data.
If there is anything else let me know.
Thanks for your interest.
Mario Lobo
--
View this message in context:
http://www.nabble.com/About-tape-usage-t1725862.html#a4697532
Sent from the Amanda - Users forum at Nabble.com.
|
