[graph-tool-tickets] [graph-tool] Ticket #156: OpenMP problem in katz and pagerank

graph-tool webmaster at skewed.de
Mon Jun 9 13:38:22 UTC 2014


#156: OpenMP problem in katz and pagerank
reopened enhancement
---------------------------------------------------------------------
Reporter: anonymous
Owner: tiago peixoto
Priority: normal
Milestone: 2.2

---------------------------------------------------------------------
Comment (by td):

(following my previous comment)

Behaviour is definitely hard to grasp.

I just tested on the 16CPU machine:

- Unshuffled: ~10s
- Shuffled: ~5.8s

I am currently running the shuffling multiple times to see if this affects the compute time significantly.

So far, what I have seen, on the 16CPU machine, running it multiple times:

- Unshuffled: stable at 9.8s
- Shuffle: Ranges from 5.8 to 6.8 seconds. Quite a significant difference (~14% difference).



In reply to your last reply:

It only includes the PR computation time. Code:

{{{
            t2 = time()
            n_iter = test(g)
            t3 = time()
}}}

With test():
{{{
def test(g):
    pr, n_iter = gt.centrality.pagerank(g, ret_iter=True, epsilon=epsilon, max_iter=max_iter)
    return n_iter
}}}

Although max_iter and epsilon are specified, they are equal to 1e-6 (default) and 100 (more than necessary).

Is there any binary of the git version? I wanted to try it but I got quite discourage by the amount of requirements (plus it looks like compilation would take a lot of time). I am currently using the Ubuntu binary.

What does the "dynamic" OpenMP scheduling does, compared to static scheduling? It increases and reduces the number of threads depending on how fast things go?


--
Ticket URL: <URL:http://graph-tool.skewed.de/ticket/156>
graph-tool <URL:http://graph-tool.skewed.de>
An efficient python module for graph analysis and manipulation.


More information about the graph-tool-tickets mailing list