Since its first announcement on June 28 here on HPCwire, and its official start on July 20, the Uber-Cloud Experiment has attracted over 160 industry and research organizations and individuals from 22 countries. They all have one goal: to jointly explore the end-to-end process of remotely accessing technical computing resources sitting in HPC centers and in the cloud. The focus of this experiment is on engineering simulations performed by small and medium enterprises that expect a quantum leap in innovation and competitiveness by using high performance computing.

The benefits of remote access to HPC are widely recognized. We have at our disposal most of the technology needed to access and run our engineering workloads on remote resources. But we still face other challenges more related to the human element. For example, trusting in the resource provider; giving away some control over our applications, data, and resources; security; provider lock-in; software licensing; unfamiliar pay-per-use computing model; and a general lack of clarity in distinguishing between hype and reality.

To explore these hurdles in detail and to learn more about this end-to-end process, we were able to build 20 teams, each consisting of an end-user and their application, the software provider, the computational resource provider, and an HPC and/or CAE expert who manages the team process. Thanks to our participants, the following teams have been established:

All 20 of these projects are underway today. Two of them are busy with defining their end-user project, 15 teams are in contact with the assigned computing resources and setting up the project environment, one is working on initiating and monitoring the end-user project execution, one is reviewing the results with the end user, and one team is already documenting the findings of the HPC as a Service process. To illustrate the team process in more detail, we present two of the projects and their current status in more detail.

Simulating new probe design for a medical device

Team Expert: Chris Dagdigian from BioTeam

The team’s end user is faced with a common problem: a periodic need for large compute capacity in order to simulate and refine potential product changes and improvements. The periodic nature of the HPC requirements means that it is not possible to have the desired amount of capacity internally as the company finds it difficult to justify capital expenditure for complex assets that may end up sitting idle for long periods of time.

To date the company has invested in a modest amount of internal HPC capacity sufficient to meet base requirements. Additional HPC resources would allow the end user to greatly expand the sensitivity of current simulations and may enable new product & design initiatives previously written off as “untestable.”

The HPC software being employed is CST Studio, a popular commercial application for electromagnetic simulations of many types. The application is currently operating in the Amazon cloud and the team has successfully completed a series of architecture refinements and scaling benchmarks. The hybrid cloud-bursting architecture allows local HPC resources residing at the end-user site to be utilized along with the Amazon cloud-based resources.

At this point in the project the team is still exploring the scaling limits of the Amazon GPU-equipped EC2 instance types and is beginning new tests and scaling runs designed to test HPC task distribution via MPI. The use of MPI will allow enable them to leverage different EC2 instance type configurations and scale beyond some technical limits imposed by the amount of memory residing within the NVIDIA GPU cards.

They believe they are currently at (or close to) the point in which they are routinely running simulations that would not be technically possible using the local-only resources of the end user. They also intend to begin testing the Amazon EC2 Spot Market, in which cloud-based assets can be obtained from an auction-like marketplace offering deeply significant cost savings over traditional on-demand hourly prices.

Multiphase flows within the cement and mineral industry

Team Expert: Ingo Seipp from science + computing ag

In this project ANSYS CFX is used to simulate a flash dryer in which hot gas is used to evaporate water from a solid. The team consists of FLSmidth as the end user, Bull as the resource provider with its extreme factory (XF) HPC on demand service, ANSYS as the software provider, and science + computing ag as team experts.

FLSmidth is the leading supplier of complete plants, equipment and services to the global minerals and cement industries. The end user needs about four to five days to complete a simulation run on the local IT infrastructure. He would like to reduce the total throughput time of the project and, in a second step, increase the mesh size to refine the results, without investing in hardware, which may not always be utilized full-time. For this, the simulation must be run on more cores and more memory through more nodes connected by a high-speed network.

XF provides 150 teraflops of computing power with InfiniBand, GPUs and currently, about 30 installed applications. Others are added on demand. Users can access XF through an easy-to-use web portal or direct login.

In this project, XF has enabled access to the end user and integrated ANSYS CFX in a web-interface for submitting jobs. For the course of this project licenses have been granted by ANSYS. The end user can manage his ANSYS licenses easily through the portal. The preparations to run the jobs are almost completed now and the first test runs should be able to start shortly.

Announcing Round Two of the Uber-Cloud Experiment

We consider Round One as proof of the concept that: yes, remote access to HPC resources works, and, there is a real need for it! And yes, there are hurdles on the way, but we know how to overcome them.

During the half-time webinar we asked the attendees if they would like to participate in the second round of the Uber-Cloud Experiment. 97 percent answered said they would. Therefore, we decided to start a new round of the experiment immediately after the first round completes. It will run from mid-November to mid-February.

Round Two of the experiment will be more professional. The end-to-end process of identifying, accessing and using remote resources (hardware, software, expertise) will become more structured, standardized, and tools-based. We will also handle more teams and more applications beyond CAE, and offer a list of additional professional services, for example, measuring the team effort. Finally, existing teams will be encouraged to use other resources, existing participants can work in new teams, and new participants can join and form new teams.

For anyone interested in learning more about the experiment or to register for Round Two, go to the Uber-Cloud Experiment website.

About the Authors

Wolfgang Gentzsch and Burak Yenier are the creators and facilitators of the Uber-Cloud Experiment. Wolfgang is an HPC veteran. Having worked in leading positions in research, academia and industry for some 30 years, Wolfgang is now an HPC consultant and the chairman of the ISC Cloud conference series for HPC and Big Data in the Cloud. Burak is the vice president of operations at CashEdge, a software-as-a-service company in Silicon Valley, which provides innovative payments and aggregation solutions to financial institutions.