| make ls print count of nodes (abrugh's idea) rather than list of nodes by default. Add -n option to print node names (never print meaningless node number). |
|
More...
|
about 15 years ago
|
| fix name collision that broke fm split |
|
More...
|
about 15 years ago
|
| Don't execute jobs owned by other users (frequently there are permisison problems) |
|
More...
|
about 15 years ago
|
| collect() should print stderr only if labelPrint=True or Verbose>0 |
|
More...
|
about 15 years ago
|
| terser output |
|
More...
|
about 15 years ago
|
| display serial, cached, and non-cached stats separately |
|
More...
|
about 15 years ago
|
| print different stats when we can tell that performance was dominated by caching. This avoids negative Wait times. |
|
More...
|
about 15 years ago
|
| bug fix in verbose output |
|
More...
|
about 15 years ago
|
| handle possible race condition where wait() beats sched() starting |
|
More...
|
about 15 years ago
|
| better verbose printing |
|
More...
|
about 15 years ago
|
| Print per-node throughput. Use a pickled dictionary for status files since it makes parsing less fragile and creates an extensible format. |
|
More...
|
about 15 years ago
|
| more informative & concise output |
|
More...
|
about 15 years ago
|
| coallesce status, stderr, and stdout collection after map into one rsync for improved latency |
|
More...
|
about 15 years ago
|
| Don't print traceback on clean exit |
|
More...
|
about 15 years ago
|
| enable %() substitution in "fm run" commands |
|
More...
|
about 15 years ago
|
| "fm _local wait" now implies "fm _local sched" so we don't have to do an extra ssh to every node |
|
More...
|
about 15 years ago
|
| fix bug in process time accounting when multiple processes per scheduler |
|
More...
|
about 15 years ago
|
| part of previous patch |
|
More...
|
about 15 years ago
|
| buffer and print labeled stderr the same way we do stdout |
|
More...
|
about 15 years ago
|
| map -f used to always use the first node. Now pick multiple nodes based on replication factor in case that one node is down. |
|
More...
|
about 15 years ago
|
| processes can be overridden per node |
|
More...
|
about 15 years ago
|
| Better SSH defaults. Add processes setting to filemap.conf |
|
More...
|
about 15 years ago
|
| print helpful error rather than asserting if -i missing on a map command |
|
More...
|
about 15 years ago
|
| improved performance reporting |
|
More...
|
about 15 years ago
|
| add labelPrint option to collect() and create "fm run" command that uses it. Fix "fm df" to also use it. |
|
More...
|
about 15 years ago
|
| specify number of processes per node in conf file. Default is 1. |
|
More...
|
about 15 years ago
|
| "map -q" option added to not cat output |
|
More...
|
about 15 years ago
|
| comment for abrugh |
|
More...
|
about 15 years ago
|
| wait() had a race that would erroneously report that scheduler died. Fixed exponential backoff in wait() loop. |
|
More...
|
about 15 years ago
|
| measure and display input bytes, wallclock time, and ratio |
|
More...
|
about 15 years ago
|
