El 18/05/2009, a las 09:49, john.mcl...@sybase.com escribió:
Having a couple thousand files or directories in a directory can be a
problem. On some systems, the more files and directories there are
in a
directory, the slower file access is, and at some point, opening a
file in
an overly large directory is a major problem. In any case, the
directories
still have to be cleaned out for re-use eventually (and to give the
hard
disk space back). The solutions that move the file to trash are OS
dependent (what function to spawn, and where to move the directory),
and
BOINC does not ever need the data back, so they should be just
deleted.
The suggestion was for using a directory called "trash" within BOINC
data directory, not the "Recycle Bin" or similar. Once the old stuff
is in the trash dir, it can be slowly deleted, while another app can
immediately use the now-clean slot.
A
better solution would be to add a "cleaning" state to the slot and to
delete only X files total from all slots every pass through the
polling
loop.
jm7
Gabor Gombas
<gomb...@sztaki.h
u> To
"Paul D. Buck"
05/18/2009 02:08 <p.d.b...@comcast.net>
AM cc
BOINC Dev Mailing List
<boinc_dev@ssl.berkeley.edu>,
Charlie Fenton
<charl...@ssl.berkeley.edu>,
john.mcl...@sybase.com, Jack
Shultz
<jshu...@hydrogenathome.org>
Subject
Re: [boinc_dev]
[boinc_alpha] New
crashs odditiy
On Fri, May 15, 2009 at 06:31:26AM -0700, Paul D. Buck wrote:
The down side to that is clean-up of completed tasks will be deferred
and slots will not be freed as fast as they are now.
Personally, I would make that trade in a heartbeat... I have lost too
many tasks to the "too many restarts" problem... and this looks to be
the real cause...
For one thing, I cannot think of a faster way to delete files on unix
systems. Does not mean that there is not a way ... but I have never
heard of it...
Hmm, how is this for simpler? When the slot is done, and ready for
clean up, do a mv to a "trash" directory mv ./1 ./trash/datetime.
Next, create a clean slot of the right number. mkdir ./1
Or change the slot numbering: instead of reusing the same number
immediately, just number them sequentially, and wrap around after a
couple thousand.
IMHO the most important thing would be to do any I/O sensitive tasks
asynchronously.
Gabor
--
---------------------------------------------------------
MTA SZTAKI Computer and Automation Research Institute
Hungarian Academy of Sciences
---------------------------------------------------------
_______________________________________________
boinc_dev mailing list
boinc_dev@ssl.berkeley.edu
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
_______________________________________________
boinc_dev mailing list
boinc_dev@ssl.berkeley.edu
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
