| patch to beat ext3's subdir limit of 32k and really to scale well beyond that |
|
More...
|
over 14 years ago
|
| make replicate and redistribute all-to-all more sane by serialzing the activities of each source node (in random destination order) so that each destination has about 1 source active at a time |
|
More...
|
over 14 years ago
|
| python 2.4 compatibility |
|
More...
|
over 14 years ago
|
| remove stray debugging output |
|
More...
|
over 14 years ago
|
| apply better sorting of nodes lists to more places |
|
More...
|
over 14 years ago
|
| Merge branch 'master' of [email protected]:mfisk/filemap |
|
More...
|
over 14 years ago
|
| improve numcmp to handle multiple transitions between numeric and non-numeric |
|
More...
|
over 14 years ago
|
| fm ls now knows if its being piped or not, useful for running wc -l and getting accurate results |
|
More...
|
over 14 years ago
|
| document new default setting |
|
More...
|
over 14 years ago
|
| add dynamicload option (on by default) to schedule more items for concurrent execution if there appears to be CPU and I/O headroom |
|
More...
|
over 14 years ago
|
| remove debugging |
|
More...
|
over 14 years ago
|
| remove dead code |
|
More...
|
over 14 years ago
|
| code to use per-process delayacct under Linux, but not commonly available, so won't use it |
|
More...
|
over 14 years ago
|
| bugfixes |
|
More...
|
over 14 years ago
|
| rename for code readability |
|
More...
|
over 14 years ago
|
| add map -qq to not print stderr results either |
|
More...
|
over 14 years ago
|
| useless code |
|
More...
|
over 14 years ago
|
| change scheduler to make local decision on whether to run an item (no output of out of date output) before bothering with global syncronization. This improves performance significantly. Also cleaned-up locking logic and removed a race condition where a node would run the same item twice at the same time |
|
More...
|
over 14 years ago
|
| expand input globs and then work our way through them without re-stating. this speeds up the scheduler loop a lot and improves child reaping times and ability to schedule lots of small items quickly |
|
More...
|
over 14 years ago
|
| rename the two finalize methods to non-conflicting names for code readability |
|
More...
|
over 14 years ago
|
| don't print text for each cpu in a node |
|
More...
|
over 14 years ago
|
| bugfix: reduce for, say, partition 7 would get all partitions that ended in 7. (leftover from when partitions were file extensions) (per abrugh) |
|
More...
|
over 14 years ago
|
| explain when/why subirs made in dst |
|
More...
|
over 14 years ago
|
| since last patch changed default relative=False, we now override with True. We want relative because we get a ton of, say, .stderr files that differ only in path |
|
More...
|
over 14 years ago
|
| fixed fm get - Don't mirror whole path (relative=False). This is what you want by default - Option parser must allow -c with only 1 arg - fm cat is now just get -c |
|
More...
|
over 14 years ago
|
| switch localpickles to make a dictionary where the key is the nodename; switch jobs to use pickles rather than get |
|
More...
|
over 14 years ago
|
| fix traceback on finalize caused by switch to non-blocking IO |
|
More...
|
over 14 years ago
|
| make option parser enforce number of fixed positional arguments |
|
More...
|
over 14 years ago
|
| replace rm_rf with shutil.rmtree per brugh |
|
More...
|
over 14 years ago
|
| merge fmplot into fm plot so we can share code. Make ls -ln use numcmp() |
|
More...
|
over 14 years ago
|
