Advanced Computing in the Age of AI | Tuesday, October 3, 2023

HPC as a Service – Key to Mainstream HPC or Wishful Thinking? 

<img style="float: left;" src="http://media2.hpcwire.com/dmr/clouds.jpg" alt="" width="95" height="71" />A recap of a blue-ribbon panel held at this year’s SRII Global Conference. The panel explored the benefits and barriers to the adoption of high performance computing and digital manufacturing with particular emphasis on “missing middle” small- to medium-sized manufacturers.


At this year’s SRII Global Conference 2012 organized by the Service Research and Innovation Institute [http://www.thesrii.org/], which aims at “Driving Research & Innovation for “IT Enabled Services” for a Better World”, we organized a panel about High Performance Technical Computing as a Service (HPCaaS).

This panel was inspired by the perception that on one hand HPC provides huge benefits to the industry, and especially to small- and medium enterprises (SMEs) with a focus on digital manufacturing (sometimes called the ‘missing middle’), but on the other hand includes a number of significant barriers preventing its wider adoption. The panel wanted to investigate how especially HPC as a Service (remote access to HPC resources, HPC in the Cloud) is able to remove these barriers. As the Council on Competitiveness [http://www.compete.org] and the National Center for Manufacturing Sciences [http://www.ncms.org] have found, 97% of manufacturers are still using workstations and PCs as their main workhorses for design and simulations, but 57% of them said that they have problems that they can't solve with their existing desktop computers today. Therefore, they have a real need for more computing power, but there are still some barriers for them to overcome.

Panel Members

To bring some light on this dilemma, panel members have been invited to reflect the complete end-to-end process of HPCaaS, starting with the SME end-user, looking for computing resources (HPC Center, HPC Cloud), software provider (ISV), and an HPC expert able to help the end-user to implement his application and data onto the resource:

-       SME HPC end-user: Frank Ding, Engineering Analysis & Technical Computing Manager at Simpson Strong-Tie, a company creating structural products that help people build safer and stronger homes and buildings.

-       HPC service provider: Ashok Krishnamurthy, Interim Co-executive Director, Ohio Supercomputer Center, known e.g. for its Blue Collar initiative.

-       Commercial HPC cloud provider: Robert Graybill, CEO and President, Nimbis Services, providing a Technical Computing Marketplace.

-       Independent Software Vendor: Nick Appleyard, Vice President - Product Delivery (PDT) Operations Americas of CD-adapco, a software company interested in moving part of his business into the Cloud.

-       HPC Expert & Digital manufacturing market analyst: Christopher Willard, Chief Research Officer Intersect360, and

-       Panel Chairman: Wolfgang Gentzsch, Executive HPC Consultant, Member of the SRII Global Leadership Team [http://www.thesrii.org/hpc-as-services] , Chairman of the ISC Cloud Conference series [http://www.isc-events.com/cloud12], and co-organizer of the Uber-Cloud HPC Experiment [http://www.hpcexperiment.com].

The Panelists’ View on HPC as a Service

SME end-user: Frank Ding, Simpson Strong-Tie

I have been an end-user of Linux cluster-based HPC for CAE since its early beginning over 10 year ago. HPC enables us to run large jobs, reduce run time, increase job throughput, and share resources among CAE analysts. It has been proven to us that HPC-based CAE is the major driver for the bottom line, i.e. product innovation, time to market, and cost reduction. I believe HPC in the cloud, or HPC as a service will be a natural evolution for the technical computing world. The immediate benefit would be breaking the HPC entry barrier for the SME, who has no in-house resources and know-how. SME can leverage the HPC expertise of the service providers without huge up-front cost. For us, who already have in-house HPC capability, I hope we can depend on HPC in the cloud in the future in case we have a surge of demand  on either HPC computing resources or application software licenses beyond what is available in-house. Currently, the bottleneck for HPC in the cloud for CAE users is internet bandwidth and ISV software licensing. I understand work is under way to establish the HPC in the cloud ecosystem. I hope the business model will work, and we will all benefit from the HPC in the cloud ecosystem.

HPC service provider: Ashok Krishnamurthy, Ohio Supercomputer Center

The Blue Collar Computing (BCC) program, launched by OSC in 2004, is aimed at bringing Modeling, Simulation and Analysis (MS&A) capabilities to the industry and business, with a particular focus on middle market companies. The large supply chain in manufacturing – the so-called missing middle in manufacturing – has been a particular focus of BCC. While the value of MS&A in improving competitiveness, reducing time-to-market  and delivering better, less expensive products and services is generally acknowledged, the missing middle nonetheless face a number of barriers in adopting MS&A. The key barriers are a lack of access to: (1) HPC resources; (2) ISV codes for MS&A; (3) expertise; and (4) education and training.

OSC’s approach is to form public-private partnerships that can collectively overcome the barriers. We work with private companies such as Nimbis Services, and non-profit organizations such as EWi and PolymerOhio who are the front-door for the companies, providing a catalog of services and being the e-commerce agent. ISV software licensing is arranged typically by the e-commerce provider or by OSC negotiating directly with the vendor. Access to expertise is from a number of sources: Engineering consulting companies, local universities and OSC staff. For education and training, OSC is working with universities and community colleges to provide general training on MS&A, while training specific to a particular software product is from the ISV.

A final issue is ease-of-access to the MS&A capabilities. OSC and Nimbis Services are jointly developing an API called the External Partner Interface (EPI) that allows an authorized entity to access OSC’s HPC resources, virtual machines and storage. Users have access to a variety of resources through EPI, from a “vertical app” such as WeldPredictor to ISV applications such as Abaqus, to a traditional Unix shell on a HPC login node. Such a unified and easy access from the desktop provides middle market companies the ability to use MS&A at their current level of capability, while providing a path for increasing usage.

HPC cloud services provider: Robert Graybill, Nimbis Services

The founding of Nimbis was motivated by three years of Council research through its HPC Initiative [www.compete.org/hpc] which shows that, while virtual prototyping enabled by HPC is a “game-changing” technology, it remains a largely underutilized national asset. Lack of virtual product simulation expertise, availability of easy to use application software, and access to more powerful systems is preventing the large and rapidly expanding group of entry-level HPC users from progressing in their use of this technology, and there is an even larger group of companies that remain stalled on the desktop.  

Nimbis, a privately held technology corporation brings an innovative technical computing marketplace solution to the cloud computing industry with a focus on the digital manufacturing sector. The Nimbis technical computing marketplace provides pre-negotiated services from compute providers, ISVs, domain experts, and regional solution providers on a “pay-as-you-go” basis for experimental and periodic users.   The solution is based on traditional web marketplace hosting, where the hosting company provides common components for building an ecommerce storefront.  The vendors (sellers) in this case are engineering services companies with application domain expertise and independent software vendors that provide consulting and support services.   In addition, the sellers may create application domain specific (vertical application) portals on Nimbis-supported compute platforms in a model similar to a smart phone “App Store”.   The intended customers (buyers) are manufacturing companies with entry-level, experimental, and/or periodic users of modeling, simulation, and analysis software. 

These characteristics of low-risk, low-effort, on-demand, "pay-as-you go" access to computing and related services enables small to midsize businesses to access easy to use “technical APPS” and/or move beyond technical computing on the desktop.

Independent software vendor: Nick Appleyard, CD-adapco

CD-adapco's PoD model (STAR-CCM+ Power-on-Demand), allows companies to access the full physics based simulation software (STAR-CCM+) on a pay-as-you-go basis running on HPC centers in the Cloud, reducing cost-to-entry for high end simulations for small to mid-size companies. A scenario would be as follows:  An engineer would like to use a physics simulation software tool to "virtually test" a product they are developing as a means to prove out its performance under an extreme range of operating conditions. Their end goal is to ensure that their product can stand-up to the warranty requirements set by a major customer they are trying to win business with. The small company does not have the upfront capital to invest in an annual license of the software and hardware necessary to simulate the corresponding physics and operating conditions  that their product must perform under. Using CD-adapco software-as-a-service model (STAR-CCM+ PoD), the engineer contacts CD-adapco to set-up a PoD account. Using a specially designed  web-based user portal for the PoD account, the engineer can then directly pay-for, monitor and manage the number of "wall clock hours" of software use that is required (similar to a phone card). The engineer, working from his/her desk to create the physics model, can run high-end simulations on any HPC provider in the Cloud that supports STAR-CCM+.

HPC expert & digital manufacturing market analyst: Christopher Willard, Intersect360

We see cloud computing as part of the logical progression in distributed computing architectures. Changes in distributed computing architectures are as much of an organizational business decision as they are a technical decision. To effect change at an organizational level requires a sound business case that covers costs, both ongoing and transitional, benefits, and risks. Because cloud computing involves transferring both data and control outside of the organization, the associated perceived and real risks make the business analysis a non-trivial activity for most organizations. User risks fall into four broad categories:

-       Security –This is both a technical problem and a “trust the provider” problem.

-       Workability – The cloud solution has to work at least as well as what it is replacing. Thus bandwidth and software are often the first concerns. Other workability issues include: system stability, network stability, system management skills and responsiveness, and so on.

-       Supplier stability – The likelihood that the vendor will remain in business, with the ability and willingness to meet its contractual obligations.

-       Supplier entanglement – The degree to which a user becomes dependent upon the vendor, and thus must accept vendor changes in pricing and technology.

Cloud service suppliers are caught in a squeeze between attempting to provide cost-competitive services and maintaining a profitable business. Broad strategies that suppliers can use include the following:

-       Monetize unused assets – Ultra-scale web service companies can offer cycles in their “capacity buffer” (i.e. the amount of overcapacity necessary to meet upcoming demand and unexpected demand peaks) as a cloud product.

-       Efficiency Play – This is a second-stage fee-for-service model that is based on cloud service providers being able to deliver computing cycles more efficiently than their customers can generate cycles in-house. Efficiencies come from a combination of economies of scale in terms of purchasing hardware, software, space, power, and so on.

-       Cross Sell – This model assumes a package of products that allows the supplier to charge a premium. The cloud computing part of the package may not be a revenue generator in and of itself, but rather an operational entry point that allows the package to work as a whole.

-       High Level Services – This model directly addresses the drawbacks to cloud computing by expanding the services provided to meet user requirements. Thus the cloud supplier provides high levels of security, ISV software license management, significant levels of storage, user specific system software environments and so on.

-       Infrastructure Taxation and Subscription – Successful “proto-cloud” business models from the internet include those used by auction sites which charge a “tax” on auctions and transactions, and gaming sites which charge subscription fees to users.

In summary, the panel confirmed that there are indeed more severe barriers for HPC in the Cloud than with mainstream cloud computing, simply because of the significant difference of the applications, web services versus big application jobs, with all their implications. But with careful analysis of each HPC application’s requirements and setting up the perfect team of industrial end-user, resource provider, ISV, and HPC expert, most of the challenges are bridgeable, which is currently explored in the Uber-Cloud Experiment [http://www.hpcexperiment.com].

Thus, there is strong indication that HPC as a Service is becoming widely recognized as a key to mainstream HPC.

EnterpriseAI