55362_en-US

liliconcom

JMP’s function to support several Distributed Computing

Hi Friends,

This is a question of JMP’s function to support several Distributed Computing.
I hope your ideas and support.

1. There are 5 nodes(80 Core) that there is CPU 16 Core in 1 node.

Can we use JMP as Distributed Computing support.

Now, MDCS(Matlab Distributed Computing Server) and MJS(Matlab Job Scheduler) was installed in header node of each computer.

After calling 16 worker in case of need by each node, they can do job scheduling and operate 5 node(80 worker)

 

2. If JMP support Distributed Computing, they think that simultaneous operations can be possible in several computers that have 16 core like MDCS/MJS. If they want to install the function, Do they need to activate what functions of JMP, or Is there any other method ?

Re: JMP’s function to support several Distributed Computing

Hi @IkjuShin,

JMP does not support a grid or distributed computing environment.

SAS has really good support for this type of infrastructure. 

Depending on the OS used in their grid environment one could deploy JMP with the proper license on one or more of the nodes in the grid.

However, each JMP instance will only use the 16 cores available on each of the single nodes at one time.

Note: Presentation server license is only supported with Windows server OS or Windows desktop OS.

--Stan

Re: JMP’s function to support several Distributed Computing

For more information on how to use the Formula Depot to deploy JMP models to a production environment, including source code for a deployment to a Spark cluster, please see https://community.jmp.com/t5/Discovery-Summit-2016/Scoring-Outside-the-Box/ta-p/22381

Re: JMP’s function to support several Distributed Computing

A couple thoughts on this:

 

1. For single computer, multi-core (i.e. desktop comptuer with 4-cores or 8-cores, like many laptops and desktops), JMP will automatically support multithreading in many places and utilize all of the available hardware. See the JMP 14 documentation for features, which support multithreading.

 

2. For many if not most of the things that JMP is typically used for, a desktop is fast enough to work with data that scientists and engineers have. In the case of the MATLAB example, you cite above, the customer is dealing with very complex simulations that take hours to complete and therefore need the overhead and hassle of setting up a distributed computing environment, scheduling jobs and farming them out to many more cores. This is also something that SAS is very good at.

 

3. If you want to "productionalize" models built in JMP, there is the option of generating score code in the Formula Depot (C, SAS, Python, JavaScript, SQL). At that point, the model can be rolled out to a system, which supports grid-like operations.