ISC Cloud 2012 BOFs: Applications/Software, Reference Architectures and Data Transfer

By Nicole Hemsoth

October 2, 2012

At ISC Cloud 2012, talking points for the Birds of a Feather sessions were hand-picked by the participants. While the importance of security was a key theme throughout the two-day event, several other salient topics emerged during the voting process. The finalized BoF roster included “Applications and software in the cloud,” “HPC Cloud Reference Architectures” and “Data Transfer in/out of Clouds” to be held in parallel. Each group had about 10-15 participants discussing the challenges and implications of their chosen topic. After the conference, the panel moderators each submitted their notes on their findings.

BOF 1: Applications/Software in the Cloud

Moderator: David Wallom, Oxford eResearch Centre

The discussion was started with the consideration of how cloud computing could change the supply of application software with the possibility of ISV partnering with cloud providers to change the delivery model. This would allow application flexibility, but it was pointed out that there is an inherent unpredictability of a pay-as-you-go (PAYG) model. It may be an issue for those groups that have been previously subjected to a fairly stable cost model, though in many other areas PAYG is becoming more normal. A problem is that in current IaaS cloud models, costing is not simple and there may be resistance to the introduction of new business models from long-term users.

It was pointed out that it isn’t just the end-user applications but also all other components. An illustration of PAYG for areas other than end-user applications, which clearly shows one of the problems with other models is where LSF is an annual license, even though it may only be required a few times (less than 10).

With this change of model, how do we support the legacy application? This will depend on the type: Community applications that are open source will have to rely on their community and commercial applications will require their users to ‘gang up’ as it were. However, there are problems with a SaaS delivery mechanism since there could be resultant legacy version support required as many commercial customers want longevity. Over the longer term, cloud migration means users will have to be more used to version migration, and if so, application providers will have to make sure version migration is easier.

The level of cloud utilization will depend on the different application communities and different maturities of software. The possibility of flexibility is strongest where software is newest, i.e., application users do not favor one model over another. It is unlikely that cloud will affect the application design model to change MPI, and thus OpenMP will still need to be supported. On a longer term, the different types of interconnection software (MPI/OpenMP) won’t matter as the hardware will catch up with newer ideas.

We mustn’t forget that software isn’t just the application but also the networks that exist around it: Community-as-a-Service and Support-as-a-Service.

Of course, less data means that it is easier to move to the cloud, but if you can do more operations on your data in the cloud then this becomes less important, for example, only downloading the important result although this may require workflow in the cloud.

With the emergence of standard APIs for different components, the time is right for application designers to accept these changes in models by moving to the most advantageous cloud provider. We must ensure that application designers learn lessons from the previous instances where public cloud providers changed their models and made previous design decisions irrelevant or less than optimal.

  • It is a whole ecosystem. Remember that:

    • The user decides on the software that best solves their problem. End users don’t care, and they just want solutions.

    • Hardware licensing versus software licensing costs can be decisive.

    • Optimization for many different types of use cases can lead to different types of hardware solutions.

    • Cloud provider chooses hardware, software, interconnects, .i.e., the most efficient solution.

    • Community clouds targeted to different communities are not inevitable but likely as different ISV and communities get together to best optimize their requirements and solutions together.

    • Whatever use of cloud or otherwise we decide on has to fit with other parts of the business model/activity.

Cloud providers have the opportunities to get away from unnecessary user complications and also support their users with new models. There are good opportunities for long term relationships between ISV and cloud providers.

Finally the difference between the cloud and Application Service Provider (which we have had for around ten years) was discussed. It was brought to light that the quality/ubiquity of the network resources and the sheer number of resource types have changed.

BOF 2: Reference Architecture

Moderator: Josh Simons, VMware

Two basic models for moving HPC workloads into a cloud environment:

1. Virtual clusters formed by creating a persistent set of virtual machines on demand. Each virtual machine runs the same software stack (OS, libraries, batch scheduler, etc.) as was used in a bare-metal environment. This is desirable because from an end-user/scientist perspective, the interface to the compute interface remains the same: they use the same batch scheduler interfaces. The use of virtual machines is transparent to the end-user.

2. Virtual machines are created on demand to run each job and they persist only for the lifetime of the job. This allows each job to run with its own custom software stack, for individual jobs to be migrated dynamically across the virtual infrastructure for load-balancing, resiliency, or power management. This is not an evolutionary model in that the end-user would need to interact with either a new software layer that understands how to launch VMs rather than scheduling onto existing cluster nodes. This could be an entirely new layer or an augmented version of existing job schedulers.

It was noted that hybrids of the above two approaches could be used as well.

The following components were identified as being critical pieces of reference architecture for HPC in the cloud. (Not an exhaustive list.)

  • Self-service capabilities to enable end-users to create clusters on the fly.

  • A catalogue of virtual machines and software stacks that can be used to create these virtual clusters.

  • A provisioning engine to instantiate these virtual machines (it was noted that Open Stack work on “placement groups” is relevant).

  • An ability to elastically flex compute resources up and down as needs change.

  • A monitoring component to watch the health and performance of the infrastructure.

  • Billing and chargeback.

  • Data staging components – to move data in and out of the cloud.

  • Policy-based resource control mechanisms to mediate access to hardware resources between multiple cloud tenants.

  • Security – data security and protection and secure isolation of workloads in a multi-tenant environment.

It was noted that a “cloud” might not be virtualized, though virtualization was seen to make a number of the above functions easier to deliver.

It was posited that once HPC moves into the cloud, there will be a need to support complex applications that require cross-cloud workflows, similar to some of the meta-computing concepts developed within the grid computing realm. It was noted that if “cloud” is the follow-on to grid computing, then it would be useful to examine grid architectures closely, to determine which features should be brought forward into mainstream cloud architectures.

There are problems still to be solved if HPC is to move into the cloud. Some are technical – end-to-end automation of the use of HPC in the cloud. Others are business related: licensing, politics, and budgetary. The budgetary issue is particularly interesting: In the face of “unlimited” compute resources, how does an organization control access to limit its budgetary spend? This is particularly important for HPC workloads, which as we know can consume all available resources at a site. What happens when such users get access to unlimited resources in the cloud? Answering these questions will likely uncover additional required components for an HPC cloud reference architecture.

BOF 3: Data Transport

Moderator: Rolf Sperber, Alcatel-Lucent

Size Matters

There has to be a differentiation concerning the size of datasets to be transported in and out of the cloud. The target is optimized access – it can be achieved for small amount of data if there is a predictable way of accessing required data or moving data in or out of the cloud. For large datasets to be transported, the quality of service will have to be guaranteed for longer periods of time.

Small Data

To have instant access to data in a cloud, current metadata will not be sufficient. A software that has knowledge of the network infrastructure and defines a virtual network on demand is required. Multiple carrier and in consequence multiple vendor environments will have to be taken into account.

Big Data

This is about huge datasets to be transported over long distances. Final target is to have predictable transfer times for multiple datasets to be transported to a single location.

First Iteration

  • Federation of folders into a single folder with a metadata server to keep track of size, locality, etc.

  • Optimize transport by means of adequate transfer software. Here we are talking about software products (most of them commercial) that help solve the TCP problem

  • Optimize access by proactive distribution if possible. Here settled paradigms of work will have to be overcome.

Second Iteration

  • Optimize transport requirements with respect to site of computation.

  • Provide network control to enable clients to define an appropriate virtual network.

    • Multiple carriers with heterogeneous environments to be taken into account.

    • Charging models to be implemented.

Third iteration or Target

  • Further optimize applications to minimize transport requirements.

  • Integrate network control into applications.

    • Federation

    • Software defined networking taking care of both dedicated instance of time when transfer starts and duration of transfer in relation to size.

    • SDN calculating both routes and time of reservation.

    • SDN calculating total duration.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion XL — were added to the benchmark suite as MLPerf continues Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire