March 24, 2020 – Charmworks will work with Argonne National Laboratory to expand and improve that Argobots threading framework with support from the Department of Energy’s Small Business Innovation Research program.
With Argobots Pro, developers can take advantage of the flexible scheduling of calculations within a node to implement powerful multitasking strategies, with a special focus on simple user-level threads. The development of the runtime system was led by the Programming Models & Runtime Systems group of Argonne and included the Urbana-Champaign of the University of Illinois. It was part of DOE’s Argo project, which aimed to create operating systems and runtimes for supercomputers in exascale format. The design of Argobots was inspired by Converse, a planning system that supports the Charm ++ parallel programming system. It was a Finalist for an R & D 100 award from R & D World magazine in 2020.
“With the expansion of multiprocessor chips and GPGPUs within compute nodes, there is a lot of asynchronous data movement and much less storage and network bandwidth. This makes the work extremely complex, dynamic and erratic, ”said Sanjay Kale, CEO of Charmworks.
“There is simply no adequate software infrastructure for work. With Argobots Pro – especially with the enhancements we’re adding to Argobots Pro – developers can combine a variety of planning strategies and schedulers that control various aspects of the hardware and applications. Whether you’re creating tasking models, runtime systems, programming language backends, or applications, you can experiment and optimize on a single platform to get the most impact. “
Some of the improvements that Charmworks and ANL are looking to implement include:
Registry-based queuing to give developers the flexibility to control sequencing across multiple queues and asynchronous events.
Node-parallel programming to improve nested parallel performance when using systems such as Open MPI or BOLT.
Mechanisms for managing communication threads to bypass bottlenecks.
A dependency tracking module for determining data readiness and coordinating data movement.
Coordinating GPGPU and accelerator kernels, monitoring completion, tracking communications and initiating data transfers.
Debugging and performance profiling tools.
Argobots Pro enables developers to better troubleshoot problems common to large-scale simulations, such as hardware failures, load imbalance, and excessive energy consumption. The runtime is highly interoperable with MPI libraries such as MPICH, Open MPI and their derivatives. It is used by many commercial code as well as open source software projects – including Intel DAOS, Margo / Mercury, and HDF5 – to achieve unprecedented performance.
Intel plans to deploy Argobots on the Aurora exascale supercomputer. Argobots is already being used on the world’s fastest supercomputer, Fugaku, and on various supercomputers in China’s national supercomputing centers in Guangzhou and Shenzhen.
Charmworks offers scalable solutions that improve parallel programming productivity. Charm ++, the company’s flagship product and technology, is an adaptive runtime and supporting tools that allow developers to easily incorporate automatic load balancing, fault tolerance, and energy conservation capabilities into their code. Charmworks also offers a suite of software based on Charm ++ including an MPI implementation called CharmMPI and a discrete event simulator called CharmDES.
Via the Argonne National Laboratory
The Argonne National Laboratory seeks solutions to pressing national problems in science and technology. Argonne is the country’s first national laboratory and conducts cutting-edge and applied scientific research in almost all scientific disciplines. Argonne Researchers work closely with researchers from hundreds of corporations, universities, and federal, state and local authorities to help them solve their specific problems, advance America’s scientific leadership, and prepare the nation for a brighter future. With employees from more than 60 nations, Argonne is administered by UChicago Argonne, LLC for the US Department of Energy’s Office of Science.
Source: Argonne National Laboratory