Tuesday 31 August 2021

Cambridge Open Exascale Lab

Cambridge Open Exascale Lab
Recently I have been occupyed writing coronavirus-related blog posts, but with the first round of the COVID-19 vaccination programme almost at its final stage in the UK, and the recovery of business activities, I plan to start writing more about spin-offs and other initiatives from Cambridge University that involve with latest innovative technologies.

In this blog post, I would like to introduce you a world-class supercomputer laboratory, Cambridge Open Exascale Lab, the University of Cambridge. It is dedicated to designing fast supercomputers to bring the UK’s science, health and industry into the levels of complexity and performance that previously were out of reach.

Cambridge Open Exascale Lab is part of the Cambridge Research Computing Services (RCS), formerly called the High Performance Computing (HPC) Facility when it was founded in 1996. Cambridge Research Computing Services was established with the aim of providing high performance computing services to leading scientists, medics and engineers across the whole of the UK.

Cambridge Research Computing Services currently runs two supercomputers, called Peta4 and Wilkes (named after Cambridge computing pioneer Sir Maurice Wilkes, 1913-2010). These are running at rates of peta (1015) floating-point operations per second (FLOPS, a measure of supercomputer performance when totalled across thousands of CPU cores, useful when handling scientific tasks that are highly parallelisable between many CPUs). The newly established Cambridge Open Exascale Lab aims to develop supercomputing systems running at the speed of exascale (1018) FLOPS, some 1,000 times the scale of the current systems.

What exascale supercomputing systems could bring
Countries from the United States to Europe to Japan are in the race to produce the world’s first exascale computing system. The United States Department of Energy and Intel announced that the first exaFLOPS supercomputer, Aurora, would be operational at Argonne National Laboratory in Lemont, Ilinois, by the end of 2021. The European Union has a range of exascale programmes in the works under its European High-Performance Computing Joint Undertaking. Japan is aiming for the exascale version of its Fugaku supercomputer to be available to users within a couple of years.1 You may wonder why exascale computing systems are so urgently being developed. Let’s explore.

Take the example of my field of expertise, biomedical science. In the last 10 years, a huge amount of biological data, such as genomic and proteomic data, has been generated, mainly due to the fall in price of sequencing. Combining this biological data with clinical data could help develop personalized medicine (also called precision medicine). Analysing an amalgamation of all this data tends to require computing systems with high processing capacity (and also high storage space, although this tends to be only a secondary consideration for supercomputers, because managing the processor interconnects is the most difficult task, whereas adding more storage is easy by comparison). With the invention of the exascale computing system, analysis of data from different data sources, which currently takes weeks or even months, could be shortened to hours or days, and thus allow more calculations to be explored within a research project’s time-frame. Personalized medicine could therefore be developed in a much faster pace.

In addition, exascale supercomputers will enable simulations that are more complex and of higher resolution. This allows researchers to explore the molecular interactions of viruses and their hosts, which aids in the design of vaccines.1 Imagine how it might have helped if we could have had a vaccine against SARS-CoV-2 with more than 95% efficacy designed in a few days, or even in a few hours, during the initial stages of the COVID-19 pandemic.

The exascale power will allow climate forecasters to swiftly run thousands of simulations, introducing tiny variations in the initial conditions, to provide better insight into the potentially disastrous effects of climate change.1

Besides life sciences, exascale computing is expected to benefit chemical design, pattern modelling, high-energy physics, materials science, oil exploration, and transportation.1

In view of the benefits of exascale computing systems, the opening of Cambridge Open Exascale Lab could help the UK remain at the forefront of different fields of science.

Goals of Cambridge Open Exascale Lab
An exascale supercomputer will contain some 135,000 GPUs and 50,000 CPUs, each one being a multi-core chip with many individual processing units. This immediately creates the problems of huge power consumption, and a potentially difficult method of programming to enable almost a billion instructions being executed simultaneously. Furthermore, if the system is upgraded, researchers may need to re-examine millions of lines of code and optimize them to make use of the extra hardware, so that the programs can reach as close to the theoretical maximum processing power as possible. In addition, the ability to access memory (RAM or long-term storage) and retrieve data quickly is an issue in highly interconnected supercomputers, as evidenced by Cambridge startup Ellexus (recently sold to Altair), which focused on profiling the bottlenecks of I/O (input and output) to the storage devices in supercomputers, and found this was frequently more of an issue than the software designers had realised.

The Cambridge Open Exascale Lab’s plans include: analyse supercomputer power consumption with a view to how to reduce it; provide a method of giving supercomputer time more quickly to scientists that need it for urgent work (such as those responding to pandemics and other disasters); apply an Intel-made programming framework that allows loads to be shared across heterogeneous computers (those involving more than one type of processor, which should make upgrades easier because any new processors do not have to be exactly matched to the existing ones); install faster storage systems (based on solid-state memory chips); improve fast communication between the parts of the computer; and work on new types of graphics to visualise the data produced by the supercomputer.

Cambridge Open Exascale Lab works with a broad range of industry, government, University and other partners. Its industry partners include Dell and Intel Corporation. With its aim to recruit 20 more staff in 2021, hopefully the lab will have sufficient support from talented people to achieve its goal in developing an exascale computing system very shortly.

Cambridge Open Exascale Lab is situated in the West Cambridge Data Centre (WCDC) built by the University of Cambridge at a cost of £20m. This centre is designed to accommodate the rapid growth in demand for high performance computing, and is one of UK’s most energy-efficient high performance computer data centres. It has a high level of security and provides research computing services at a national level.



Most information written in this blog post is mainly from the website of Cambridge Open Exascale Lab.

References:
1. Adam Mann. Core Concept: Nascent exascale supercomputers offer promise, present challenges. PNAS, September 15, 2020 117 (37) 22623-22625).

No comments:

Post a Comment