Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > comp.soft-sys.math.mathematica > #11657 > unrolled thread
| Started by | tredmond59@gmail.com |
|---|---|
| First post | 2012-09-07 09:02 +0000 |
| Last post | 2012-09-12 07:03 +0000 |
| Articles | 5 — 3 participants |
Back to article view | Back to comp.soft-sys.math.mathematica
LaunchKernels[] and Java tredmond59@gmail.com - 2012-09-07 09:02 +0000
Re: LaunchKernels[] and Java David Bailey <dave@removedbailey.co.uk> - 2012-09-08 07:48 +0000
Re: LaunchKernels[] and Java tredmond59@gmail.com - 2012-09-12 07:29 +0000
Re: LaunchKernels[] and Java David Reiss <dbreiss@gmail.com> - 2012-09-10 08:13 +0000
Re: LaunchKernels[] and Java tredmond59@gmail.com - 2012-09-12 07:03 +0000
| From | tredmond59@gmail.com |
|---|---|
| Date | 2012-09-07 09:02 +0000 |
| Subject | LaunchKernels[] and Java |
| Message-ID | <k2cd6d$f5s$1@smc.vnet.net> |
I have a machine with 8 cores and I have been getting amazing results with the parallel computation. LaunchKernels[] gives me 4 kernels and many of my problems actually are probably running something like four times as fast. This is really awesome! But I am wondering if I can get Java to connect to all of these kernels. I have some java code that I invoke from the mathematica user interface. This program will often run overnight or even for more than a day. I figured if Java will talk to all these kernels then I could divide the problem and it would run much faster. (Conceptually the problem splits up nicely.) Does anyone know how to do this? -Timothy
[toc] | [next] | [standalone]
| From | David Bailey <dave@removedbailey.co.uk> |
|---|---|
| Date | 2012-09-08 07:48 +0000 |
| Message-ID | <k2et7q$km5$1@smc.vnet.net> |
| In reply to | #11657 |
On 07/09/2012 10:02, tredmond59@gmail.com wrote: > I have a machine with 8 cores and I have been getting amazing results with the parallel computation. LaunchKernels[] gives me 4 kernels and many of my problems actually are probably running something like four times as fast. This is really awesome! > > But I am wondering if I can get Java to connect to all of these kernels. I have some java code that I invoke from the mathematica user interface. This program will often run overnight or even for more than a day. I figured if Java will talk to all these kernels then I could divide the problem and it would run much faster. (Conceptually the problem splits up nicely.) > > Does anyone know how to do this? > > -Timothy > I suspect the 4 kernel limit may be determined by your license. I am pretty sure if you use J/Link in 4 kernels, you will get 4 copies of Java running. I'd communicate to Java from one kernel, but then use threads within the Java to spread the load across the available CPUs. David Bailey http://www.dbaileyconsultancy.co.uk
[toc] | [prev] | [next] | [standalone]
| From | tredmond59@gmail.com |
|---|---|
| Date | 2012-09-12 07:29 +0000 |
| Message-ID | <k2pdlf$9g5$1@smc.vnet.net> |
| In reply to | #11673 |
> I am pretty sure if you use J/Link in 4 kernels, you will get 4 copies > > of Java running. I have done a bit of work and this is pretty much what I have running now. Right now these four java copies are running independently. This is extremely good but not ideal. Today I had a new idea which I will try shortly. I was thinking that if I use RMI to make these four java's communicate, I can have one multi-threaded java that has full access to all four kernels. In fact I am wondering if I can get access to five running kernels - I might be able to also push the master kernel which usually just co-ordinates the work of the other ones. Thanks for the help - I will let you know if this works.
[toc] | [prev] | [next] | [standalone]
| From | David Reiss <dbreiss@gmail.com> |
|---|---|
| Date | 2012-09-10 08:13 +0000 |
| Message-ID | <k2k7ei$12g$1@smc.vnet.net> |
| In reply to | #11657 |
Execute $ConfiguredKernels to find out how many Kernels are available on your system. ?$ConfiguredKernels $ConfiguredKernels is the default list of kernels that are configured to be launched for parallel computing. >> The Documentation Center states, for $ConfiguredKernels, "On a multicore computer the list typically includes as many local kernels as there are cores, up to a limit defined by license availability." --David On Sep 7, 5:03 am, tredmon...@gmail.com wrote: > I have a machine with 8 cores and I have been getting amazing results wit= h the parallel computation. LaunchKernels[] gives me 4 kernels and many = of my problems actually are probably running something like four times as f= ast. This is really awesome! > > But I am wondering if I can get Java to connect to all of these kernels. = I have some java code that I invoke from the mathematica user interface.= This program will often run overnight or even for more than a day. I figur= ed if Java will talk to all these kernels then I could divide the problem a= nd it would run much faster. (Conceptually the problem splits up nicely.= ) > > Does anyone know how to do this? > > -Timothy
[toc] | [prev] | [next] | [standalone]
| From | tredmond59@gmail.com |
|---|---|
| Date | 2012-09-12 07:03 +0000 |
| Message-ID | <k2pc3g$99j$1@smc.vnet.net> |
| In reply to | #11687 |
> > The Documentation Center states, for $ConfiguredKernels, "On a > > multicore computer the list typically includes as many local kernels > > as there are cores, up to a limit defined by license availability." > I think that this is right - I have 8 cores but I am limited to four kernels by my license. -Timothy
[toc] | [prev] | [standalone]
Back to top | Article view | comp.soft-sys.math.mathematica
csiph-web