Large physical models and detailed analyses are pushing the limits of the available in-house hardware resources, and engineering companies are running more simulations than ever as part of their design cycle. With advantages in hardware accessibility, cost and ease of use, cloud-based solutions are gaining traction in the field of computer-aided engineering (CAE) and more specifically in computational fluid dynamics (CFD) (1). Flow Science, in collaboration with Penguin Computing, is introducing a powerful cloud-based solution that provides an end-to-end use of its CFD software FLOW-3D. This article presents an overview of the various features and highlights the ease of use and flexibility of FLOW-3D in the cloud.
Cloud Service Provider – Penguin Computing
Deploying and managing effective high-performance computing (HPC) environments can be daunting and expensive (2). Rapidly changing hardware, administration duties, performance and capacity are issues that HPC solutions need to contend with regularly. Penguin Computing offers an industry-leading cloud-based solution that delivers outstanding usability, performance, security and cost effectiveness through their Penguin Computing on Demand (POD) platform. POD allows users to utilize a bare-metal HPC environment in the cloud without having to invest in on-premise infrastructure while providing top-notch security for users’ data. The bare-metal hardware provided by Penguin Computing includes the latest processors and InfiniBand interconnect, making it an ideal platform for FLOW-3D. In Q3 2016, Penguin Computing introduced their 3D-accelerated remote desktop solution called Scyld Cloud Workstation (SCW) which provides true multi-user desktop collaboration for cloud-based desktops. SCW provides secure, easy remote access to Linux workstation-class desktops through standard, modern web-browsers while eliminating the need for client-side installations. SCW offers significant time savings by removing the requirement to download large data files, and moving pre- and post-processing to the cloud for highly efficient workflow (3). The SCW platform also provides access to the compute resources hosted by POD.
FLOW-3D Integration on POD
FLOW-3D’s HPC version hosted on POD SCW has shown scaling of up to hundreds of cores for real world simulations. Flow Science has developed a robust graphical user interface (GUI) for FLOW-3D that enables the seamless usage of POD cloud resources from the SCW platform. With an emphasis on usability, while retaining all the features of FLOW-3D’s desktop version, the FLOW-3D POD GUI allows users to submit and monitor simulations by accessing the POD job submission queue system. Post-processing through the GUI and FlowSight can be accomplished in real-time with the desktop version with little or no noticeable delays in graphics processing.
Below is a video tutorial of how to run a simulation using the FLOW-3D POD GUI. This tutorial emphasizes the key features of this new GUI.
This video demonstrates the responsiveness of the graphics on POD SCW. It emphasizes the little to no lag time in the toggle operation of the graphics window on POD SCW.
The POD Run Options pop-up in the FLOW-3D GUI allows users to configure the settings for a simulation to be run on POD. The pop-up parameters are used to create job submission scripts for the automatic decomposition tool (ADT) and the solver. Once these scripts are created, the GUI submits these jobs and monitors their progress for the user. This image shows the default settings for the pop-up, and the table below summarizes each of these settings.
POD Run Options pop-up for FLOW-3D POD GUI
|Queue||T30 – Haswell
H30 – Sandy Bridge
M40 – Westmere
|POD queue that will be used to run a simulation. Once chosen, the queue is fixed for the ADT and solver step.|
|# of cores||Defaults – 20 (T30), 16 (H30) and 12 (M40)||Number of cores requested for a given simulation. This is always a multiple of 20 for T30, 16 for H30 and 12 for M40.|
|Wall clock time requested:
|Defaults: 1 min||The estimated run time for decomposition and simulation step. This parameter is required by the POD queue system to prioritize a job. It is recommended to over-estimate these times to ensure the completion of a job. User will only be charged for the duration of the simulation. Longer wall clock time requests may result in longer wait times for the job to start.|
|Failsafe termination||Default – 10 min||This option allows the simulation job to be terminated smoothly when the wall clock time approaches the end of wall clock time requested. The default behavior only initiates a restart edit 10 min before the end of the wall clock time requested. When checked and time ‘n’ minutes specified, the GUI initiates a termination at ‘n’ minutes before the wall clock time requested.|
|Decomposition directions||Default – All three directions||Unchecking a direction prevents decomposition in unchecked direction for ADT.|
Using FLOW-3D on POD and through SCW can be accomplished in two ways. If a user already has a FLOW-3D license locally and would like to go to POD for big jobs that only come up once in a while, he or she can use FLOW-3D on POD in a pay-per-use mode. As an example, running a job on 100 cores for 24 hours (big job) would cost about $1,000 to $2,000 in FLOW-3D license charges. With this option, there are no limitations in the number of simulation instances or number of cores each simulation can run on.
Another mode of operation is for the user to acquire his or her FLOW-3D license ahead of time and host it on POD. In this case, since the licensing fee has already been settled, there will not be any additional license charges for using POD. However, in this scenario, the user would be limited to the number of instances and cores of the pre-paid license.
In all cases, there are nominal hardware charges for CPU and storage usage on POD, as well as for renting an SCW. CPU rates are $0.1 per core-hour for the latest Haswell chips (T30) with 20 cores within a node, and $0.09 per core-hour for the previous generation Sandy Bridge chips (H30) and Westmere chips (M40) with 16 cores and 12 cores respectively within each node. Storage is free for the first GB and $0.1 per GB per month after that. An SCW is a 6 core virtual workstation with about 30 GB of effective memory and excellent NVIDIA graphical capabilities, and can be rented for $150/month.
All in all, this service is affordable and fits well within Flow Science’s existing licensing models. It can also be used on a pay-per-use basis for the peak times during a design cycle. For more information, contact email@example.com.
- Waterman, Pamela J. The Sky is the Limit for Cloud Computing. [http://www.digitaleng.news/de/cloud-based-analysis-silver-lining-or-white-fluff/] s.l. : Digital Engineering, 2015.
- POD, Penguin Computing. POD HPC Cloud. [http://www.penguincomputing.com/solutions/pod-hpc-cloud/]
- Penguin Computing. Range of Upgrades Focus on Quality of Service and Security Enhancements. [http://www.penguincomputing.com/company/press-releases/penguin-computing-multi-user-remote-desktop-for-scyld-cloud-workstation/] Fremont : Penguin Computing, 2016.