Effective but stingy installation of MDCS
Show older comments
If there are 3 nodes and 12 cores each, mdcs has to be installed in all 36 cores right?
Now for the nodes to communicate with each other, is installing mdcs in 1 core (or the master) of each node enough for the nodes to communicate with each other? Since we can use PCT to work with the cores within the system, we don't need to install MDCS in every core right?
This might mean a compromise on speed but is it possible?
Answers (1)
Kojiro Saito
on 16 Dec 2017
Edited: Kojiro Saito
on 16 Dec 2017
0 votes
MDCS is to be installed in each node not per core. You can use one MDCS worker per node which utilizes one CPU core and also launch multiple workers (for example, 12 workers which utilizes 12 cores) as much as the maximum number of your MDCS worker license.
9 Comments
Viswanath Hariharan
on 18 Dec 2017
Kojiro Saito
on 19 Dec 2017
No, MDCS licenses are purchasable per a worker but minimum workers are 16 workers. Here is MDCS price link.
Viswanath Hariharan
on 21 Dec 2017
Kojiro Saito
on 22 Dec 2017
Technically, total 9 workers (per a worker in each system) will work for communication and they will use one CPU core in each system. In terms of licenses, a minimum license volume is 16 workers, so you need to purchase at lease 16 workers licenses.
Viswanath Hariharan
on 5 Feb 2018
Viswanath Hariharan
on 5 Feb 2018
Kojiro Saito
on 6 Feb 2018
Yes, with MDCS 16 workers licenses, you can install MDCS in 9 nodes, and launches 1 worker per node, then you can make 9 nodes communications with 9 workers. Each worker consumes 1 CPU cores, so only 1 core is used per node.
Before purchasing, I recommend that you would test it with a trial license. You can submit a trial MDCS license from the link below.
Viswanath Hariharan
on 6 Feb 2018
Jason Ross
on 16 Feb 2018
The default is that each worker has one computational thread. You can change that in the cluster profile to allow the use of more computational threads per worker.
Categories
Find more on MATLAB Parallel Server in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!