On Wed, Jul 01, 2009 at 12:28:34PM -0400, Chris Hoogendyk wrote:
>
>
> Paul Bijnens wrote:
> >On 2009-06-30 19:52, Chris Hoogendyk wrote:
> >>So, the question is, if Amanda has more than one holding disk
> >>(partition), and they differ in size, will Amanda know when the
> >>smaller one is inadequate for a particular DLE and explicitly choose
> >>the larger one? Also, if I have specified spindle numbers in my
> >>disklist, so that Amanda will avoid doing parallel dumps from the
> >>same spindle, is there any way of informing Amanda of the spindle
> >>numbers for the holding disks (partitions) and taking that into
> >>account in the planning?
> >
> >No problem at all.
> >
> >You make config entry for each holdingdisk.
> >On each holdingdisk you can specify peculiarities of the disk,
> >like the path to the toplevel holdingdirectory, and the amount of
> >free space that should not be used etc.
> >
> >Make sure to specify a chunksize that will fit easily on the smallest
> >disk. I make my chunksize 1GB.
> >
> >Amanda will spread the holdingdisk data over all the holdingdisks
> >avoiding any problems when one DLE would not fit on a single area.
> >
> >As side benefit you'll also get some improvement in throughput because
> >now taper reading finished dump images will compete less with the
> >dumpers writing to disk.
> >
> >Making it different filesystems instead of one large logical volume makes
> >future adding/removing/swapping disks easier as well. And having
> >a disk error on one of the disks in raid0 lvm's is much worse
> >than on independent filesystems.
>
> Also acknowledging Brian's comments.
>
> A1: So, Amanda will make use of all the holding space available, and can
> go across holding disks with a DLE if one fills up. That answers the
> practical question regarding it simply working.
> Q2: I presume then, since Paul & Brian have spoken up, and no one has
> mentioned it, that Amanda has no understanding of spindle numbers with
> regard to holding disk space. So, it will just be chance that determines
> whether a DLE, that happens to be on the same spindle as one of the
> holding disk partitions, gets dumped to that particular holding disk
> partition. Of course, this is then the disadvantage of using a partition
> on a shared disk rather than dedicating the whole disk for holding
> space. And, on those systems where I have the luxury, I have configured
> two whole disks as two separate holding disk spaces.
I have not read, and do not see an option in amanda.conf, for spindle
number to come into play for the holding areas, so I assume you are
correct. Yes, dedicated spindles are prefereable to shared in this case.
I beleive that amanda utilized the holding area starting with the
first listed and progresses upwards. I have assumed they are used
strinctly in a round-robin fashion, but I have never tested/investigated.
Since the dumps are run in a seemingly/relatively random fashion I
don't know of a way to schedule around the issue you are describing.
> Q3: This does still leave a choice. If I can only manage one disk's
> worth of space, should I go through the effort of rearranging existing
> partitions in order to shuffle space and get one dedicated disk for
> holding disk space? The question is whether one dedicated disk will lead
> to better efficiency than multiple holding partitions spread across a
> few shared disk drives (where a DLE and a holding space may be on the
> same spindle). I would have all dumps going to the same spindle, but it
> would never be the same spindle that a DLE was being dumped from.
The only time a single large spindle would be required if you wanted
to restore a complete DLE from tape to the holding area (and the given
DLE was larger than any available holding area) to pick it apart for
some reason. for instance, I find it simple enough to pipe amrestore
to ufsrestore, but have difficulty trying to pipe amrestore through
gtar to pick specific files/directories, so in those cases I may
restore the DLE completely. You can avoid this particular problem
by using amrecover (I believe, I need to experiment more).
> And, for those who mentioned it -- No, I won't be using LVM. For one
> thing, this isn't Linux. For another, it is a legacy system with lots of
> drives and partitions that have all been assigned for their own
> particular uses and needs. On my newer systems, I'm using ZFS (which
> obsoletes other volume managers), but I'm still keeping Amanda holding
> disks as separate dedicated disk drives with UFS. This guarantees no
> issues with i/o contention or bandwidth.
>
>
> --
> ---------------
>
> Chris Hoogendyk
>
> -
> O__ ---- Systems Administrator
> c/ /'_ --- Biology & Geology Departments
> (*) \(*) -- 140 Morrill Science Center
> ~~~~~~~~~~ - University of Massachusetts, Amherst
>
> <hoogendyk AT bio.umass DOT edu>
>
> ---------------
>
> Erd�s 4
>
>
---
Brian R Cuttler brian.cuttler AT wadsworth DOT org
Computer Systems Support (v) 518 486-1697
Wadsworth Center (f) 518 473-6384
NYS Department of Health Help Desk 518 473-0773
IMPORTANT NOTICE: This e-mail and any attachments may contain
confidential or sensitive information which is, or may be, legally
privileged or otherwise protected by law from further disclosure. It
is intended only for the addressee. If you received this in error or
from someone who was not authorized to send it to you, please do not
distribute, copy or use it or any attachments. Please notify the
sender immediately by reply e-mail and delete this from your
system. Thank you for your cooperation.
|
