My main use case for Serverless is to run simulation workloads currently. A quick specific example of a simulation workload I'm running with Serverless is that we have a product which runs circuit simulations. This can traditionally run on HPC clusters, but we are also working on running it on a serverless architecture for these simulation jobs. We are currently in the POC area and will spend more time to understand how our customers can use this architecture to achieve better turnaround time for their simulations instead of going the traditional HPC schedule-driven approach.

The best features Serverless offers for my use case are specific things within the simulation where we call it optimization. Batch simulation does not work out for us because this is more of an interactive session wherein a thick client can start an iteration and we need remote processing for the actual optimization. Once the optimization result for a single optimization gets back, the thick client can tune it and send out the next optimization. This is the major thing we are trying to achieve using Serverless.

This has changed the workflow and experience for my users, which is a big plus because the optimization jobs are very latency sensitive. We cannot do this using a traditional HPC scheduler because of how HPC schedulers work. We are looking at Serverless for quick scheduling of these jobs, and the turnaround time for optimization has increased tremendously. This is a big win.

I do not have any specific suggestion at this point on how Serverless can be improved for my use case. Perhaps down the line I will have more information.

The reason I choose a seven is primarily resource-based scheduling. This is what we are not able to get around on Serverless. It solves a main use case, but it still does not solve our batch executing and batch simulation use cases. However, if this could be one solution to move away from a traditional HPC scheduler, I would have definitely given a ten.

I have been using Serverless for just about three to four months.

Serverless is stable in my experience so far.

Serverless's scalability is something that since we have very specific needs, we are not looking at scalability at this point. However, down the line, that is going to be really important if we do a production deployment wherein we expect maybe tens to hundreds of users using it at the same time. That is when the scalability will be a big challenge. At this point, we do not have an issue with scalability.

The customer support experience has not been applicable to me, as this was done by my solutions architect, so I never interacted with the customer support.

We did not previously use a different solution, as we are a traditional HPC batch scheduler.

We did not previously use a different solution, as we are a traditional HPC batch scheduler.

We did not previously use a different solution, as we are a traditional HPC batch scheduler.

I have seen a return on investment so far, as I can definitely see a cost saving when it comes to keeping a virtual machine running. Serverless actually saves money for our POC. I can definitely say that. However, it is too early to say a comparison between how much we are saving from a traditional scheduler to Serverless computing.

My experience with pricing, setup cost, and licensing is that some of these are done by my solutions architect and I will not have enough information about that in my current role as a product manager.

Before choosing Serverless, I did not evaluate other options, as we are still working with Serverless. We have not looked at other solutions.

The transition from HPC clusters to a serverless architecture is going so far by facing the biggest difference where the traditional HPC schedulers have quite a bit of customization when it comes to scheduling jobs. It is traditionally a batch execution. Going for Serverless, we are looking at far quicker scheduling of jobs, but we currently are facing some issues when it comes to using the resources for the jobs and getting the image loaded because the image is very huge.

It is too early to tell if I have any numbers or specific metrics on how much the turnaround time has improved since moving towards Serverless. We are still working on optimization of this entire workflow. We are still trying to figure out all the use cases we can work with. We still need to benchmark the results, so it is too early to provide that information.

The advice I would give to others looking into using Serverless is to especially focus on any traditional HPC application, try to modularize your application and figure out specific workflows which are well suited for Serverless and migrate those into Serverless. I definitely see cost saving, quick scheduling of jobs and the overall improvement in efficiency. I gave this review a rating of seven out of ten.