save-aggregates-copy-sequence

I cross-posted this message on the mailing list and the usenet. I hope
not to annoy anybody.


At our departement we were testing pine on a distributed filesystem
with bad results.
It required 30 secs to move 1000 messages between two mboxes of 10000
msgs size (about 150Mb).

The inner issue was the fs slowdown, but we were wondering why only
pine had such a bad performance.
We discovered that the problem was in a visualisation option: "reverse
sort of messages".

Looking at strace output we noticed that with this option active, each
message was handled separately. I mean that each message required
* opening destination mbox
* locking destination mbox
* copyng the message
* fsync'ing (!!!!) the destination mbox
* closing the destination mbox

fsync is too costly on our fs.

One of our staff dicovered a hidden feature in the source code of
pine: save-aggregates-copy-sequence
With this feature active the copy process works quickly with only one
fsync at the end even when messages are reverse sorted.

We are now wondering why isn't this feature added as a default one. Is
there some problematic context where this feature slows down
operations or produces unconsistent state of mailbox?

Thanks,
  Matteo Faleschini
_______________________________________________
Pine-info mailing list
[email protected]
http://mailman1.u.washington.edu/mailman/listinfo/pine-info