On Fri, Jun 02, 2006 at 09:14:13AM -0700, Joe Donner (sent by Nabble.com) wrote:
> 
> Or does it work this way:

You are fairly close.

> 
> Amanda sends compressed data to the tape drive.

If talking about amtapetype, it sends random data, uncompressed.

> 
> The tape drive also compresses the data, and therefore actually expands it. 
> Amanda doesn't know this.

Not also, but it does expand it.  And yes, amanda doesn't know this.
It only measures what it sends.  Keep this in mind below.

> 
> After Amanda has sent 135GB to the tape drive, the tape drive has actually
> written 160GB to the tape, and tells Amanda that it's full.

right
> 
> Amand therefore thinks the tape drive's capacity is only 135GB.
> 
> Is that correct?

yup.

> 
> So if I use this tapetype definition, will Amanda in future only ever send
> 135GB to the drive?

nope.

> 
> If I disable hardware compression, then I will always get a minimum of 135GB
> worth of capacity.  If I modify the tapetype definition to contain, for
> example 155000 mbytes, then I will get a minimum of approximately 150GB
> worth of capacity, but depending on how compressable the data is, it may in
> fact be a lot more than that.
> 
> Does that make sense?

Could, but not quite what happens.  The capacity of the tape is used mostly
in planning and in deciding what completed dump in the holding disk to send
next.  If you are using the "largest fit" taper algorithm, then an accurate
capacity is needed.

But, once taper starts to send a completed dump to the tape, it runs to its
full capacity, not consulting the tapetype definition as to when to stop.

> 
> Furthermore, if Amanda can be configured to send uncompressed data to the
> drive, and the drive has hardware compression on, then I could expect a
> capacity of around 320GB, depending on how compressable the data is?

The decision to use SW or HW compression is a matter of amanda knowing, with
some measure of accuracy, how much tape capacity is available.  If you use
HW compression, amanda never knows for sure.

Also, some DLEs may contain uncompressible, thus expanding data with
HW compression.  With SW comp. you can chose to not compress that DLE.

However, SW compression takes time and cpu cycles.  Lots of cpu cycles.
These cycles can be spent on the client or on the server, which ever has
the greater abundance.  And if done on the client, the network load is
decreased because less data is transmitted over the network.

> 
> My head hurts...

Stop beating it against the wall :)

Learn how to turn off HW compression.  Set amanda up to use SW compression.
If the cpu burden is too much, you can later easily switch to HW.  At that
time you will have some data as to the compressability of your data and
you can fudge a tapetype capacity based on that.

-- 
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)