Re: [X10-users] confused about the performance

Hi, With Fib, memory allocation garbage collection dominate the execution time. The runtime creates objects internally for each async in your code. You probably also create a Fib object for each computation step. The performance of the JVM GC in this context is much better than the one of the

Re: [X10-users] confused about the performance

Hi, Two comments on your performance issues with Fib.x10 1. Looking at your Fib code, I see you are still allocating objects on the sequential leafs of your computation. For r small you really want to compute Fib with something like: static def seq(r:Int) { if (r 2) return 0;

Re: [X10-users] getAvailableProcessors?

Is there a way in X10 to query the number of physical processors at a Place? Something analagous to Runtime.getRuntime().availableProcessors() in Java? Not at this point. Olivier -- This SF.net email is sponsored by

[X10-users] New plugin dependencies - update site

If you are working on the x10 compiler and runtime source code in eclipse you will notice that starting with revision 18897 we added plugin dependencies to the x10.wala project. The required plugins can be obtained from the eclipse update site: http://dist.codehaus.org/x10/wala Olivier

Re: [X10-users] X10_NTHREADS slowdown with IO

Hi Sascha, I am hoping to look at your code on Thursday. But as others have reported IO performance is currently not good. Also, we have been working on fixing the thread busy waiting issue but the code is not complete yet. I may have something for you to test in a few weeks but there is no

Re: [X10-users] X10 Runtime in X10

Hi Sascha, Your understanding of async is correct but finish is not implemented this way. The activity blocked on a finish is not pushed on the queue. The queues always contain activities that have not started yet, never suspended activity. When an activity blocks on a finish, it calls

Re: [X10-users] X10 Runtime in X10

, wheather 2011/8/22 Olivier Tardieu tard...@us.ibm.com Hi Sascha, Your understanding of async is correct but finish is not implemented this way. The activity blocked on a finish is not pushed on the queue. The queues always contain activities that have not started yet, never suspended

Re: [X10-users] More threads than specified in X10_NTHREADS.

Hi, We support thread affinity via the X10RT_CPUMAP env variable. But I think the setting is only supported for x10rt pami. Moreover this only gives control over X10_NTHREADS. Any other thread dynamically added by the runtime is free to roam. Therefore X10RT_CPUMAP is only intended to be used in

Re: [X10-users] Clock bug?

Von: Olivier Tardieu [tard...@us.ibm.com] Gesendet: Freitag, 4. Mai 2012 21:02 An: Mailing list for users of the X10 programming language Betreff: Re: [X10-users] Clock bug? Hi, I think you are expecting outputs from distinct places to show in order

Re: [X10-users] FT-alltoall Benchmark, libpoe.so, libmpi_ibm.so

Hi The makefile for this benchmark is platform-specific and assumes pami transport. Try compiling with: x10c++ -O -NO_CHECKS -STATIC_CHECKS fft.x10 -o fft The code itself does not require pami or mpi. It will run fine with sockets. The only hack is the Random class, which replaces the portable

[X10-users] X10 submission wins the HPC Challenge Best Performance Award in the Class II competition

makes its decision based on a report, code for benchmarks, and a presentation. The submission was authored by Olivier Tardieu, David Grove, Bard Bloom, David Cunningham, Benjamin Herta, Prabhanjan Kambadur, Vijay Saraswat, Avraham Shinnar, Mikio Takeuchi and Mandana Vaziri of IBM Research

Re: [X10-users] remote async idiom

Hi Josh, The spec is outdated. at (p) async is preferred to async at (p). In the first case, p is evaluated synchronously (i.e., by the parent async) and then a single remote async is spawned. In the second, p is semantically required to be evaluated asynchronously with the parent async as it

[X10-users] X10 2.4 outlook

Folks, As you know we released X10 2.3.1 last week. We have now started working on our next release: X10 2.4. For 2.4 we are planning a limited but significant revision of the core standard library. This is going to involve a series of disruptive changes to trunk. So unless you are willing to

Re: [X10-users] Clocked async question

Hi, There are really 3 activities registered with the clock: not only A and B but also the parent activity while in the body of the clocked finish. If A is spawned first and reaches advanceAll before B is spawned then A has to wait on the parent activity. The parent activity only releases the

Re: [X10-users] Inquiring about how to properly x10's work-stealing scheduler

of work- stealing, then work-stealing scheduler itself will have nothing to do. Please let me know how should I implement this. I will be thankful for your help. Thanks and Regards Sparsh Mittal On Wed, Apr 17, 2013 at 2:56 PM, Olivier Tardieu tard...@us.ibm.com wrote: Hi

Re: [X10-users] Embed option

Hi, The @Embed annotation was an experiment. It is not currently maintained and does not do anything. Instances of the annotation are still floating around as we might in the future make it work again. The intend of this annotation is to permit to embedding an object instance into another

Re: [X10-users] Resilient X10, async, X10_STATIC_THREADS and DeadPlaceException

-3368)? Cheers and thanks in advance, Marco [attachment DPEExample.x10 deleted by Olivier Tardieu/Watson/IBM] -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests

Re: [X10-users] Resilient X10, async, X10_STATIC_THREADS and DeadPlaceException

to fix the problem. Am I doing something fundamentally wrong or is this a bug? Might it be related to Bug #3368 (http://jira.codehaus.org/browse/XTENLANG- 3368)? Cheers and thanks in advance, Marco [attachment DPEExample.x10 deleted by Olivier Tardieu/Watson/IBM

Re: [X10-users] GLB and X10_STATIC_THREADS

Jonas, X10_STATIC_THREADS is a Boolean variable. 1 and true are equivalent settings. Our convention is that a Boolean environment variable that is set to something other than false or f or 0 is considered set to true. Olivier Jonas P jopo...@gmail.com wrote on 05/24/2014 07:13:59 AM: From:

Re: [X10-users] distributed APGAS 1.0

Hi, This is not easily done in the APGAS 1.0 release but this is possible if you download and build apgas from the github repository. The launch syntax is the following: java -Dapgas.launcher=apgas.impl.SshLauncher -Dapgas.hostfile=/home/tardieu/hostfile -Dapgas.places=4 HelloWorld This

Re: [X10-users] APGAS lib: Adding Places across different mashines

her question about the SshLauncher: When I have allocated eg. 5 > Nodes in a Cluster with 2 CPUS each, then I want to start 10 places > overall, 2 places at every Node. Is there a way to do this easy with > parameters? > > greets > Jonas > > 2015-09-30 16:38 GMT+02:00 Oliv

Re: [X10-users] APGAS lib: Adding Places across different mashines

Jonas, I pushed new launchers to git. I added a new ssh launcher. java -Dapgas.launcher=apgas.impl.SshLauncher -Dapgas.hostfile=/home/tardieu/hostfile -Dapgas.places=4 ... It uses ssh to run on remote nodes. Place 0 always run on localhost. The hostfile should contain the hosts for the

Re: [X10-users] APGAS lib: Adding Places across different mashines

Hi Jonas, Are you trying the trunk or the 1.0 release? The 1.0 release has issues with name resolution and with nodes with multiple network interfaces. I recently made some changes to the trunk to try to fix these issues. You should use the trunk if you are still on the 1.0 release. With

Re: [X10-users] [APGAS] Questions about ResilientUTS

Hi Jonas, About DeadPlacesException at line 386. I am not sure. What exception other than a DeadPlaceException are you seeing inside the MultipleException? About data loss. When I started working with Hazelcast, there was no reliable way to detect data loss. Hazelcast has added an API for data

Re: [X10-users] [APGAS] Questions about ResilientUTS

o not > understand the mind of MultipleExceptions. When should a > MultipleException be thrown and when a DeadPlaceException? > > > Best wishes, > Jonas > > Am 09.07.2016 um 15:08 schrieb Olivier Tardieu: > > Hi Jonas, > > > > About DeadPlacesExceptio

Re: [X10-users] [APGAS] Performance at place failure

re there other options to improve APGAS's performance at a place failure? > > > > Thanks and cheers > > -- > Jonas Posner > Universitaet Kassel > Fachbereich 16 Elektrotechnik/Informatik > Fachgebiet Programmiersprachen/-methodik > Wilhelmshoeher Allee 71-

Re: [X10-users] [APGAS] Performance at place failure

Jonas, I have now reduced the connection retry count to 0. Olivier Olivier Tardieu/Watson/IBM@IBMUS wrote on 10/03/2016 09:30:03 PM: > From: Olivier Tardieu/Watson/IBM@IBMUS > To: Mailing list for users of the X10 programming language us...@lists.sourceforge.net> > Date: 10/03/2

Re: [X10-users] [APGAS] Unexpected NullPointerException from an at in an placeFailureHandler

Hi Jonas, The place failure handling scheme is a work in progress. In principle I would like place failure handlers to be a kind of optional task (that runs iff a failure is detected) inside the enclosing finish. In other words, the finish should not complete before the handler returns but the

Re: [X10-users] [APGAS] Performance at place failure

oreover, if I use hazelcast transactions in a APGAS program, I am using > additionally this: > TransactionOptions txOptions = new TransactionOptions().setTimeout(10, > TimeUnit.SECONDS); > > Otherwise, hazelcast waits in worst case 120 sec for a transaction. > > > Thanks and cheers > Jonas &g