This is a preliminary page on testing.  For now, there is just one test, which is a cpu/memory ’sanity check’.  If you observe that ALLPATHS-LG runs slower than expected, we encourage you to run this test.  It will help to determine whether the problem is with ALLPATHS-LG or instead with computational infrastructure.

1. ParallelSortTest

This module is included with the ALLPATHS-LG distribution.

The command:

ParallelSortTest X=4


- Do not run this on a machine having less than 64 GB of RAM.  It might crash such a machine.
- If you run this on a huge shared machine, you should add “P=n” where n is some sensible number of processors.  The default is to use all processors.

The output:

It reports the amount of time used in ParallelSort.  This is what we benchmark.  Please note that the total time for the test is substantially larger.

Benchmark values

(based on three successive iterations, essentially no other activity on machines):

Time (sec) Cores Memory (Gb) Company Product Processor Broad Machine
20-23 48 512 Dell PowerEdge R815 AMD6174/2.2 crd9
69-89 32 256 HP ProLiant DL785 G5 AMD8454/2.2 crd4
100-109 16 128 Dell PowerEdge R905 AMD8350/1.0 crd2
117-134 8 64 Dell PowerEdge 6950 AMD8216/2.4 wga8


(a) If you run it and get much worse results, it may be that something is wrong with your
computational infrastructure.
(b) We have observed that in some cases, a machine can get into a funky state, resulting in poor results from this test, but after rebooting, results return to normal.
(c) If you get much better results than we do, please share your results.  We and others will be interested.

Comments are closed.