On Wed, Jan 04, 2006 at 04:13:25AM -0600, Hommersom, Gerrit (G) wrote:
> I've followed with great interest the discussion on this topic as it
> pertains to my backup system
>
> The problems with premature end-of-tape failures persist with big DLE's.
> (order 30-35 GB). The problem gets more severe from 1 failure per cycle to
> daily failures currently. No dumps are left on the holding disk.
>
> Closer inspection of the Exabyte FAQ learned that this may occur if the
> datarate to the drive is too low. The drive waits for input while writing
> 'gap tracks' that uses physical space.
>
> I re-examined the amanda logs and saw that for large DLE's the datarate for
> the dumper and taper are the same and in the order of 3300 KB/s. VXA2 units
> write at 6000 KB/s
> The main raid-5 taperack is an external SCSI2 unit. The tape is external
> SCSI2 too. The holding disk is 39 GB on an internal SCSI2 disk.
>
> My questions:
> will amanda use the holding disk for DLE's from on local disk to another
> i.e. will DLE's from the external filsystem use the holding disk before
> taping.
Basically the amanda server, when backing up itself, is backing up
a remote client. It doesn't do anything unique for "local" DLEs
vs. "remote" DLEs. They are all remote.
> I accidentally defined a holdingdisk chunksize of 10 GB on a standard (32
> bit) linux system. What would be the effect: direct write to tape ??
No, chunksize is the "maximum" size of any file it will create on
the holding disk. If a DLE is larger than 1 chunk, additional
chunks are used. Where this would create problems is if your OS
did not allow files larger than 2GB (used to be pretty common) and
amanda tried to create large chunks. But that would be a different
problem than not using the holding disk.
> What factors may affect the datarate to the tape unit ??
I suspect you have already identified your problem area. Your
holding disk is not being used, either not at all, or at least
not for your large DLEs. I suspect this because of the similarity
of taping and dumping rates (how fast do other large, but not
humongous DLEs tape?) and the close match in size of the DLEs
causing problems and the total holding disk space available.
Amanda decides on a DLE by DLE basis weather to dump to holding
disk or directly to tape. Part of its decision tree is the
available space on holding disk. The amount available is generally
lower than the total because some is "reserve"d for incrementals
when the tape drive is unavailable (see your amanda.conf setting,
reserve defaults to 100% reserved) and by the amount allocated
to dumps of other DLEs in progress.
So, for a 30+GB DLE, the space available on a 39GB holding disk
could easily be too small thus causing direct to tape dumping.
So, check if other big DLEs seem to be using the holding disk,
check your reserve parameter and lower it, consider decreasing
the size of your DLEs, increase the size and/or number of hold-
ing disks (you can define multiple). They need not be dedicated.
I think I have 4 holding disks defined. They are just directories
on file systems that generally have a decent amount of free space.
I then tell amanda it is free to use all the space on that FS
"except" a certain amount (syntax is something like 'use -2G')
and I define a chunksize reasonable for the amount of space
on that holding disk, big space - large chunks, small space -
small chunks.
Let us know how you make out.
--
Jon H. LaBadie jon AT jgcomp DOT com
JG Computing
4455 Province Line Road (609) 252-0159
Princeton, NJ 08540-4322 (609) 683-7220 (fax)
|
