On Tue, May 29, 2001 at 05:04:33PM -0500, John R. Jackson wrote:
> What do you mean by "over [the] network"? What were you talking to on
> the other side? A tape drive? Or writing into a file?
We have experimented with both sending the backups directly to tape and
writing them to the server's local volume. The bottleneck with TAR seems
to be traversal through the network. Too small a block size simply
results in slow throughput -- whatever the receiving end is.
> It won't care at all because it will soak up the stdout of the gtar
> process and rework it into 32 KByte blocks for transfer over the network
> (and storage in the holding disk and tape).
So in essence, tweaking the blocking factor won't make a difference in
terms of throughput, since Amanda will use 32kB regardless of what was
specified to TAR. I think this will be "good enough" as I haven't yet
had the time to compare the throughput of TAR with 32KB and 63kB.
> My guess would be you'd add a new flag to the dumptype amanda.conf parsing
> (server-src/conffile.c) and then pass that along to the client in the
> OPTIONS string. Just trace how "exclude" is handled and you should get
> the general idea. It's not all that complicated.
>
> Note that the code to pass --atime-preserve is already in Amanda, but it's
> a compile time rather than run time option.
I'll have to look into it. Thanks for turning my head in the right
direction. This might actually be a good candidate for a run-time option.
;-)
Lari
--
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
| Lari Huttunen | +358 50 338 0759 | http://www.helsinki.fi/~lshuttun |
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
