Each year the cloud faithful converge on Las Vegas in the fall for AWS re:Invent. This year’s event delivered exciting announcements for ANSYS users interested in performing engineering simulation on the AWS cloud.
With well over 30,000 attendees, the 2016 conference was too big to be contained within the expansive Venetian hotel/Sands expo complex and it spilled across into adjacent facilities (comfortable shoes were a requirement). Wednesday’s keynote session by Andy Jassy, CEO and Thursday’s session by Werner Vogels, CTO highlighted the growing reach of AWS. The conference featured a staggering number of new features, services and some powerful new hardware.
There were more new launches this year than I could possibly cover in one blog post, but there were more than a few announcements worth highlighting for ANSYS users interested in using the public cloud for engineering simulation. These new features will no doubt power upcoming enhancements to ANSYS Enterprise Cloud
New instance types will push cloud HPC boundaries
One of the key advantages of the public cloud is the rapid hardware refresh rate. Each year brings new, more powerful virtual machines and this year was no exception. Amazon announced the general availability of the new R4 memory-optimized family of instances. At the high end, these instances double the memory, number of vCPUs and network bandwidth of the previous generation R3 instance family. ANSYS Enterprise Cloud currently uses R3’s for both high-memory interactive sessions and for the auto-scaling HPC cluster used for Mechanical simulations. Watch for an upcoming blog post on updated Mechanical solver benchmarks!
Also announced were the next-generation compute-optimized instances, the C5 family. Based on Intel’s Xeon “Skylake” processor, C5’s feature up to 72 vCPUs and 144 GiB of memory per node and ENA networking which delivers up to 20 Gbps. ANSYS Enterprise Cloud currently uses C4 instances for the auto-scaling HPC clusters used for CFD and Electronics batch solves. When the C5 instances become available, it will be interesting to update our CFD solver benchmark tests to see how high this improved performance will push the ceiling of HPC on AWS.
Powerful new features for cloud-based graphics
To take full advantage of HPC in the cloud, it’s important to support more than batch compute workloads. To enable the full engineering simulation workflow, we need to be able to work with the results in the cloud. Given that simulation workloads include complex 3D graphics, that necessitates a server-side GPU for good graphics performance. The challenge is getting a virtual machine that include the right mix of GPU, CPU and RAM.
For ANSYS Enterprise Cloud, we engineered a solution that combines a high-memory virtual machine with a separate virtual machine that includes a GPU in order to deliver a high-powered cloud-based graphics workstation experience; for that we rely on NICE Software’s external rendering feature for DCV. AWS acquired NICE Software in early 2016 and has used their technology to deliver Amazon EC2 Elastic GPUs, which allows you to attach a GPU (several size options are available) to the virtual machine of your choice.
“AWS works closely with technology leaders like ANSYS to ensure that we’re delivering the functionality needed to help customers achieve their goals. To get the full benefit of cloud-based HPC, ANSYS recognized that engineering simulation users require access to interactive sessions with both GPU-accelerated graphics AND larger amounts of RAM and made this as a cornerstone capability of ANSYS Enterprise Cloud. We’re pleased that Amazon EC2 Elastic GPUs will make this ability available to all AWS customers.”
Matt Garman, Vice President, Amazon Elastic Compute Cloud (EC2)
Also announced was AppStream 2.0; the second generation of AWS’ application streaming service which will make it easier for ISV’s like ANSYS to deliver rich cloud-based graphics experiences right in your web browser. I was proud to join Tosh Tambe and Supreeth Sheshadri of AWS on stage at re:Invent as a launch partner during the break-out session highlighting Elastic GPU and AppStream 2.0.
Here’s the full video.
An emerging batch service
Another announcement highlighted in Andy Jassy’s keynote session was a new batch job service called, sensibly, AWS Batch. While it’s still in preview mode, this fully managed service promises to greatly simplify the process of orchestrating large-scale HPC jobs on AWS. We’ll be keeping our eye on this service as it becomes generally available and later adds support for MPI parallel jobs.
Putting it all together
So if you combine new high-performance instance types with enhanced high-bandwidth low latency networking, the ability to attach a GPU to any instance types and new services for streaming graphics sessions and managing batch jobs, it’s easy to see that AWS is getting very very serious about supporting cloud-based engineering simulation at scale. These new launches will improve performance, reduce complexity, improve user experience and drive down the cost of performing engineering simulation on the cloud.
And then… having a little fun
AWS also announced new services aimed at bringing Artificial Intelligence into the mainstream. Amazon Lex brings the engine that powers Amazon Alexa to enable developers to build conversational interfaces using voice and text (GE has already announced a lamp you can talk to).
“Alexa, run a design optimization over the weekend on spot instances only if the market price stays below 10 percent of the on demand price and show me the results which most closely resemble…”
Incidentally, the conference swag this year included an Echo Dot for every attendee, so clearly Amazon wants me to be ready.