I would like to know that what is the root cause for this testing in moldflow as below,
1. Run four cases via parallel computing just need 6hr16min to completed for each case, that is to say the total elapsed time for four cases are 6hr16min.
2. Run four cases via one by one computing need 18hr46min to completed for four cases, that is to say the elapsed time for each case is 4hr37min.
That question is that why the analysis time four cases via parallel computing is shorter than four cases via one by one?
What is the theory foundation for parallel computing and one by one computing in moldflow?
Well I guess your computer have more than one core so it can do many things in parallel simultaenusly. All main Moldflow solvers are parallelilized that mean they can speedup the analysis by using multiple cores. Still the speedup is usually less than linear meaning that during analysis some cores are just waiting for their coworkers to finish what they are doing. Multiple analysis are independent from each other and the processor cores could do their job without waiting for each other (so the speedup is almost linear) - unless we hit a limitation by memory.
Lets have an analogy - suppose you have a family of eight and you need to assemble a few pieces IKEA cupboards. You can all work on one cabinet then work on the second one, etc. Some work would be done truly in parallel - e.g. if 80 nails is need to be driven into the backpanel of the cupboard, each family member would drive only ten and the work will be done 8 times as fast as one worker would do it, other work would (e.g. putting on a dorknob) would be done only by one worker and the rest of the family would just wait for him to finish. In the result the cupboard will be assembled faster than one worker would do it but not 8 times as fast as by one worker). If you have a large room and enough cupboards the 8 people could assemble their own cupboards in parallel, they would not wait for each other and they would do the work 8 time as fast as a single worker. Thus, if they assemble their cupboards in parallel they would do the work faster than if they all work on a single pieace of furniture than on the next one, etc. Finally if your models are big and there is not enough memory for them all some memory would be swapped to the harddrive. It would be equivalent to assembling cupboards in a small room, there some pieces are moved to the storage and back dorung assembling - obviously it would slow down the process.
The bottom line is that if you have many analysises to run then the fastest way is to have one analysis per processor core unless the models are big and you hit the RAM limit. If you often limited by computer power you may consider to run analyses on a cloud rather than locally. Autodesk cloud has an almost unlimited number of cores.
This is very interesting question. We are considering to deploy a private cloud for CAE tools including Moldflow. May I know if you have recommended scheduling strategy for Moldflow simulation using Cloud?
Do we need to recommend user to submit each scenario one by one into Cloud? Or they can submit a simulation job with a lot of scenarioes into Cloud together? The former way needs a lot of license, and simulation can not be fast if only using one core per scenario.
The latter way can save license, but I don't know if it is feasbile to submit such kind of job into Cloud by MSDOS command to request for parallel computing for a lot of scenarioes at same time.
Do we have best practice to manage moldflow job in Cloud?
What did you set for your number of threads in the advanced option ?
Also, it is a bit confusing what you mean by parallel, if you speak about the number of cores you are using, or the number of job you were running simulatenously.
Global Engineering Systems
Sent from mobile phone
Also, do you have a GPU on your machine ?
In fact, I do agree, it is not expected behavior. So the configuration of machine, the running processes have most likely impacted the solvers. I would suggest you talk to support.