Article
Technology
Into the light - the power of optical HPC
22 March 2016
With the bioinformatics market expected worth to be £13bn, and the genomics market estimated at over $22bn[1], both by 2020, there is an “unquenchable demand for HPC," according to Chairman of Optalysys James Duez.
The rise of high-performance computing (HPC) responds to the ever-increasing demand for more efficient and powerful processing devices to help solve some of the world’s greatest problems.
This is where we come in – to enable our computational biology research, our scientists need to process an enormous amount of biological data facilitated by HPC.
A 4300% growth in annual data generation by 2020 is predicted[2]. With such a considerable increase in data, the problems of energy efficiency, high expectancy and the rising cost of compute resources will only become more pressing.
Energy footprint.
EI and the HPC hardware provider Optalysys joined forces to introduce the ground-breaking optical processing device GENESYS to perform large-scale DNA sequence searches for crucial genomics research powered by just a mains supply, and aims to be one of the top 50 fastest processing devices in the world by 2017.
Some of the largest high-performance computers cost millions of dollars so they are not cheap. Performance and scale are typically the main drivers for HPC; the trade-off is usually more cost for better speed and scale. This is where optical computing is hoping to break the trend.
Our HPC systems collectively consume up to 130KW of power, including mechanically removing heat. Powered by optical light, GENESYS is expected to reduce our energy consumption by over 95% - significantly reducing the environmental impact of running traditional HPC.
HPC systems consume vast amounts of power and generate significant heat e.g. the world's fastest supercomputer the Tianhe-2 uses 24MW of power[3] and costs $21m/year to run[4]. A comparable optical ‘supercomputer’ based on Optalysys technology will run from a standard mains supply, consuming, at least, four orders of magnitude less power.
A EI Benchmark BLAST search on a single HPC node takes 28 hrs with a measured energy footprint of 11.2 kWh (including cooling). From this, with the optical processor, it is estimated that an annual cost saving will be over £40k.
We speak to James Duez from Optalysys to see how the revolutionary technology is tackling big data …
World breakthrough.
What are the constraints of current HPC technology for processing current big data?
Many organisations require large high-performance computers to perform mathematically complex calculations, especially science and engineering institutions. Certain mathematical processes do not scale well on HPCs and it takes increasingly large numbers of processors to work together to meet demands. Often results can take hours, days or even weeks to complete. The amount of big data we are working with is also increasing rapidly making the challenge even harder. The world is looking for a breakthrough and a step-change.
Is the Optalysys third generation optical processing technology the “step-change”?
he Optalysys technology takes a completely different approach and uses light instead of electricity to process the data. This means that our hardware is very fast but with low energy consumption.
We encode the data we are working with into a beam of light so we can search through large libraries of data rapidly. Within a genomics use case, we will be able to perform BLAST-type searches faster and more cost-effectively that with a traditional HPC. Processing time should reduce from 28 hours to under 90 mins or perhaps even less.
As part of the project, we hope to have a 1.3 petaflop equivalent optical system by 2017 powered from a standard mains supply. This size of machine currently takes megawatts with traditional HPC. The largest system in the world currently is about 33 petaflops and there is only a handful of petaflop systems worldwide.
The Optalysys processor is also very small relative to the speed it can deliver, being around the size of a desktop computer. This raises the prospect of bioinformaticians being able to have their own personal HPC.
Leading Institute.
Since EI became a partner in the optical processor what progress has been made with project GENESYS? Why did you choose EI?
The project has been going very well, and we have already demonstrated our third-generation prototype to over 60 bioinformaticians at the Norwich Research Park. We are soon to demonstrate our fourth-generation hardware which, once the electronics and software are complete, will be capable of meeting or even exceeding the requirements of our collaborative project.
As a leading centre for genome analysis in world, EI has a great reputation in the scientific community – by needing supercomputing power to tackle some chunky biological challenges, our interests were aligned and the partnership seemed like a favourable choice.
Why do you think optical processing has taken off so much for the bioinformatics community?
I think it’s the prospect of each bioinformatician having their own desktop-sized super-computer sitting on their desk, one that can complete complex searches in a few minutes rather than weeks. That will enable more time to be spent on the science, and less queueing for HPC resources.
Speed of light.
Does the technology apply to other industries’ super computing capabilities?
Absolutely. We are currently focusing on two main mathematical processes: optical correlation, which is most relevant to genomics, and spectral derivatives which are used in Computational Fluid Dynamics applications like weather prediction. The processor is capable of being used in many industries including medical scanning, oil and energy, finance analysis, signal processing - to name but a few.
What are the challenges you have come up against with using light rather than electricity?
Of course, the processor still uses electricity to operate but uses low-power laser light for the actual data processing which is inherently low energy and unlike traditional processors, does not require cooling.
Our past technical challenges have come from the very sensitive alignment tolerances required, and the need to eliminate noise from the system. Our proprietary designs eliminate these issues.
Our current challenges are not so much about proving principles which are now well understood, but rather scaling our technology to (and beyond) the target speeds. This requires further electronics development. Of course, we have more work to do but it is going well so far.
Optical 'Black box'.
You mentioned the technology as a desktop sequencer without the need for a supercomputer – how will this work?
The Optalysys device is best visualised as a board, similar in size and shape to that of a high-end graphics card which will be connected to a host computer on a desk, or mounted in an HPC rack. Inside the box, the optical components themselves involve a laser, camera and liquid crystal microdisplays bonded to a pure glass block. This is mounted on some electronics and of course, the device will have embedded software.
Both reference and search data are passed by the host computer to the device for alignment - much in the same way as they are passed to an HPC now. The difference is, the searches are carried out optically and the results are passed back to the operator in the same format as they are now. The Optalysys device will really be a 'black box’ that can drop into existing pipelines so that Bioinformaticians do not need to develop new ways of operating.
The Bioinformatics code that will call our new device (via API) will be open source, enabling developers to make modifications and find new ways of using this exciting new hardware. We hope that scientists and engineers will take this code and find entirely new things for our processor to do!
Aerodynamic processing.
What’s next for the Optalysys-EI partnership?
I hope we will find other project work to extend bioinformatics applications for this exciting new technology beyond the BLAST test cases that are defined in our project plan. There is a lot of scope for us to transform applications within many areas of bioinformatics.
Do you have future plans for the technology to make further impact in the HPC community?
We have many plans and are already working in other collaborations including a Horizon 2020 project (ESCAPE) as part of an EU-wide consortium led by the European Centre for Medium-Range Weather Forecasts (ECMWF). In that project, we are working alongside others in the HPC community to explore how optical computing can revolutionise weather prediction and lead to exaflop equivalent speeds. We are also working with partners in aerodynamic research.
We ask our EI GENESYS Lead Dr Tim Stitt what he thinks of the project so far:
“Researchers will continue to be dependent on HPC systems to help analyse the large amounts of data that is being generated by modern sequencing platforms. Whether you leverage cloud computing resources or procure, host and maintain your own local HPC resources, the cost can escalate quickly, particularly for research groups with small computing budgets.
“With the advent of Optalysys’s optical computing technology, researchers will now have the opportunity to host a GENESYS system on their own desktop, with the equivalent performance of a Petaflop supercomputer for sequence alignments, solely powered from a local mains supply and for an overall cost much less than current cloud and HPC services.
“We are very proud to be in partnership with Optalysys to help realise this goal and be at the forefront of advancing computing capability within the Life Sciences.”