MPI Is Not Perfect … Yet

By Daniel Holmes

August 3, 2016

The Message Passing Interface (MPI) is the standard definition of a communication API that has underpinned traditional HPC for decades. The message passing programming represents distributed-memory hardware architectures using processes that send messages to each other. When first standardised in 1993-4, MPI was a major step forward from the many proprietary, system-dependent, and semantically different message-passing libraries that came before it.

However, software engineering has moved on since the 1990s, and so have computer hardware, languages, and mass-market expectations. Is MPI still relevant? Do we need something else to augment it or to replace it?

Last year, Jonathan Dursi wrote a popular blog post entitled “HPC is dying and MPI is killing it”.

Rather than dying, I believe HPC is expanding, evolving, and fragmenting. And MPI can, and should, play a significant role in nurturing the new HPC into being.

Dursi had a few misconceptions on MPI but also some very valid and valuable insight. During the months after that blog post, Dursi interacted with Jeff Squyres; an active member of the MPI Forum and core developer of the Open MPI library. Squyres challenged him on various points and asked for further information on his perceptions of MPI. Dursi has written several follow-up blog posts, both positive and negative towards MPI (http://www.dursi.ca).

For me, these exchanges highlight the need to involve the wider community in the development of MPI. As an active member of the MPI Forum myself, and having written an implementation of the MPI API in C# for my PhD, I need that perspective from outside the already-converted MPI community. We often respond to the stuff below MPI (hardware and system software) but rarely to the stuff above MPI, to which we need to start paying more attention. There is a lot of useful information in those blogs and the discussions that happened offline. There is general appeal that the community should be aware of and partake in. This discussion needs to continue.

I want MPI to be challenged. Especially at the MPI Forum and the EuroMPI conference, the focus is often on how good MPI is: all the problems that are perceived with it are not really MPI’s fault. If only people knew more about to use MPI properly, then they would be singing its praises too. But MPI is not actually perfect – why would the MPI Forum meet four times a year if it were? Why do 20 organisations pay to send representatives? What are we talking about? Why do we still need the EuroMPI conference series?

This year I am the Programme Chair for EuroMPI 2016, so I should have an answer for at least the last of those questions!

I set the theme for this year’s conference as “Modern Challenges to MPI’s Dominance in HPC” and built a programme that allows the MPI community to begin to identify, and to meet, those challenges.

I invited Jonathan Dursi as a keynote speaker because, through that blog, he put his head above the parapet and started a discussion that we urgently need to have. To broaden the ‘from outside MPI’ perspective, I invited experienced technology journalist Nicole Hemsoth.

I am relying on Bill Gropp’s keynote talk to defend MPI – MPI was, is, and always will be, a fantastic thing, both good and necessary for HPC. Both the MPI Standard and the MPI library implementations, embody a wealth of knowledge, experience, thought, and innovation over decades, including before it was called MPI. It would be a huge disservice, and a big mistake, to just throw it away.

Kathryn Mohror, chair of the Tools working group in the MPI Forum, and David Lecomber, CEO of Allinea Software Ltd, focus on one of the most user-oriented aspects of MPI, its support for tools that aid development, debugging, and optimisation.

The key discussion I want to take place is between the establishment and the radicals, on what MPI is right now, and the different perspectives on what is needed for the future.

Why doesn’t Google use MPI for search, or Facebook in their datacentres? Why isn’t MPI used in high frequency trading? These are all things where high performance communication and extreme scale matter.

We heard at ISC this year that the AI/Deep Learning/Machine Learning community have discovered MPI and are starting to use it and rely on it. Why did this take so long? Why have they only just discovered it? It has been around since the 90s. Is it really the best tool for the job or just the closest thing to hand?

We need these opinions to be heard so that MPI can address the current and future challenges of high-end computing.

MPI is already changing.  Every year there are optimisations of existing functions. Every year we talk about tweaks to the API to make it slightly better, more understandable, or less ambiguous. But the MPI Forum is also talking about three much bigger new concepts: endpoints, fault-tolerance and sessions.

MPI Endpoints is a reaction to the increasing number of cores and threads in a shared memory node (below MPI) but also to the increased usage of shared memory programming models in concert with MPI (same level as MPI). Fault tolerance is a response to the decreasing mean time before failure with the ever-increasing system scale (below MPI) but also a realisation that MPI cannot just take the whole system down at the first fault when there are other components that can continue (above MPI). MPI Sessions is about managing MPI’s interaction with increasingly complex runtimes (below MPI), but is also a response to modern programming techniques, such as modularisation using libraries (above MPI).

As with many previous changes in MPI these changes respond to increasing complexity and capability of hardware and system-level software. They are responses to the stuff below MPI, but they are also mindful of the stuff above MPI.

I hope that EuroMPI this year will be a crucible for these ideas, a meeting-point between MPI’s past and its future.

Everyone who is involved in HPC should have an interest in the outcome.

I hope that the discussions held at this meeting will influence the path that MPI takes and it is vital that everyone has a voice in this.

About EuroMPI

http://www.eurompi2016.ed.ac.uk/

Now in its 23rd year, EuroMPI is the leading conference for users and developers of MPI. The aim of this conference is to bring together all of the stakeholders involved in developments and applications related to the Message Passing Interface (MPI). As the preeminent meeting for users, developers and researchers to interact and discuss new developments and applications of message-passing parallel computing, the annual meeting has a long, rich tradition. This year’s conference will be held in Edinburgh, Scotland, UK from 25th to 28th September 2016.

There is still an opportunity to contribute to the technical program content by submitting a poster by 26th August 2016. The organisers particularly encourage those who use MPI and work with it every-day but would not otherwise consider themselves part of the traditional EuroMPI community to submit as we wish to hear this group’s experience of MPI.

About the Author

An Applications Consultant in HPC Research at the EPCC, The University of Edinburgh, Daniel Holmes’ research focuses on the message passing interface (MPI) and interoperability with other programming models and systems. His work involves implementing MPI libraries in a variety of different languages, designing and implementing new features for effective message passing and improving basic parallelism strategies in a range of situations from shared memory to PGAS systems. He is an active member of the MPI Forum, chairing a working group and chapter committee. Prior to his research career Daniel gained twelve years experience as an IT professional in a wide variety of companies and industries.

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