Re: [Duplicity-talk] Version 0.6.0 Released - Checkpoint/Restart

[Kenneth Loafman]

Peter Schuller wrote:
>> Makes sense. Unless you say otherwise before I have time tonight I'll
>> make the change, try to test in on sigfiles that exceed the size of
>> each read and submit it, hopefully saving you some effort.
> 
> https://code.launchpad.net/~scode/duplicity/reasonable-io-blocksize
> 
> Merge request generated.
> 
> The changes were conservative in that they only changed the block
> size. I tested the result both with and without --no-encryption, and
> with signature files a few times larger than the 128k block size (on a
> ~ 100 meb backup). The resulting backup was successfully restore and
> diff -uNr:ed.
> 
> Kenneth, is there a reason for having a *minimum* block size? Do small
> requests cause any particular problem with one of the expected source
> iterators?
> 
> In addition, do we really need to maintain such exact volume sizes
> (even to the point of topping it off)?
> 
> My thinking is that if we can eliminate both "special" treatments, the
> implementation would be dead simple and idiomatic for streaming I/O (a
> regular "read until EOF or bytes > target" type of thing).

I know of no reason for minimum block size except that extremely small
block sizes cause high IO overhead.  As long as the minimum is something
reasonable there is little overhead.  Of course, if we were to properly
tune this, we'd look at file system block sizes, alignment, and the
whole nine yards.  Maybe in the future.

As to topping it off... I think there was a bug in gpg that caused this
problem.  Remember, duplicity has been around since 2002, so a number of
workarounds may still be in the code that are not needed.

BTW, looked at the code and its clean.  Thanks for the patch.

I'm not sure what's going on with Launchpad.  I've been trying to wrest
control of the trunk away from vcs-imports for the last week, but some
idiot keeps hijacking the change request with nonsense that has nothing
to do with the request itself and causing it to be marked answered.  I
then have to remind them that the original request has not been
answered.  One way or the other I'll get it released.

...Thanks,
...Ken

signature.asc
Description: OpenPGP digital signature