* Jon LaBadie <jon AT jgcomp DOT com> [20070720 09:53]:
> On Fri, Jul 20, 2007 at 07:16:21AM -0400, Jean-Francois Malouin wrote:
> > * Olivier Nicole <on AT cs.ait.ac DOT th> [20070720 00:51]:
> > > > A DLE using tape spanning from a client (amanda 2.5.1p3 Debian/Sarge)
> > > > to server (2.5.2p1 SGI/Irix) just seems to dump **extremely** slowly:
> > > > so far in ~8hrs not even one chunk of 9GB has made it to tape. All the
> > > > other DLEs (non-spanning) on that client dump as fast as the network
> > > > allows and all the other DLEs (local, with or without spanning) dump
> > > > as fast as the tape drive can stream. 
> > > 
> 
> I don't recall the details, but when I first set up tape spanning I ran
> into problems with system settings for shared memory segment size.  This
> limited the maximum spanning chunk to some tiny amount and caused the
> number of chunks to go into the hundreds.  As each chunk is a separate
> tape file, and the tape drive stops and restarts between each tape file,
> the result was just like a drive that was shoe-shining rather than
> streaming.
> 
> Check your system and amanda logs for messages regarding shared mem.

Debug files and systems logs don't show anything that would help at
first glance.

On DLEs local to the server I see good performance when going
DLE -> shared segment -> tape. I tried with tape_splitsize values
of 4GB and 9GB and I can see no difference as far as tape performance.
What could be the difference when the byte stream comes from a remote
client? Obviously apart from the fact that it's not local :)

Would it help to turn on some debug_chunker or debug_taper?

thanks jf


> 
> -- 
> 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)

-- 
<° ><