OK !

We are working on the American eel genome and I found the issue while comparing 
two runs with same
parameters.

On 12/13/2012 12:08 PM, Louis Letourneau wrote:
> Adrian had more issues than I did. I now rarely need checkpoints since my job 
> finish in a timely manner with X thousands of cores now.

What's the value of X ?

The best performance I obtain on large super computers is with
the polytope for routing and with a diameter of 2.


For instance:

Vertices: 1024 = 32*32
Type: Polytope
Degree: 62 = (32-1)*2
Diameter: 2 (32^2=1024)
Options:  -route-messages -connection-type polytope -routing-graph-degree 62

or

Vertices: 4225 = 65*65
Type: Polytope
Degree: 128 = (65-1)*2
Diameter: 2 (65^2=1024)
Options:  -route-messages -connection-type polytope -routing-graph-degree 4225


With OpenMPI, I use

   -bind-to-core -by-node


-Sébastien

>
> It's nice that you found the issue though.
>
> Louis
> ________________________________________
> From: Sébastien Boisvert [[email protected]]
> Sent: Thursday, December 13, 2012 12:02 PM
> To: Louis Letourneau
> Cc: [email protected]
> Subject: Regression in the checkpointing system
>
> Hello Louis,
>
>
> A few months ago, you told me that Ray v1.7 was better than both Ray 2.0.0 
> and Ray 2.1.0 for
> your work. I also remember that you are using the checkpointing in Ray.
>
> I just found out that the checkpointing in Ray is broken since 2.0.0 for the 
> extension of seeds.
>
> So any job that starts from seeds that are in checkpoints won't produce any 
> extension and
> contigs will therefore be short.
>
> I will cook a patch for this.
>
>
>
> -Sébastien
>


------------------------------------------------------------------------------
LogMeIn Rescue: Anywhere, Anytime Remote support for IT. Free Trial
Remotely access PCs and mobile devices and provide instant support
Improve your efficiency, and focus on delivering more value-add services
Discover what IT Professionals Know. Rescue delivers
http://p.sf.net/sfu/logmein_12329d2d
_______________________________________________
Denovoassembler-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/denovoassembler-users

Reply via email to