The Lanka Education and Research Network (LEARN) is embarking on a mission to make high-performance computing (HPC) more accessible to researchers in Sri Lanka. This move is envisioned to set the foundation to overcome the pressing need among local researchers for greater computational power and data storage. According to Consultant CTO of LEARN, Dr. Asitha Bandaranayake, the organisation aims to introduce HPC in a series of planned steps, including the establishment of a state-of-the-art data centre in the country, specifically targeting the research and education sector. “LEARN has observed the trend towards heightened demand for computing capabilities and storage. Over recent years, this demand has intensified, not only with the rise of AI and deep learning technologies but also with the emergence of data-intensive fields like Biotechnology, Climate Science, Smart Agriculture, etc. Concurrently, there’s a growing appreciation for converged solutions, although the expertise required to implement and sustain such virtual cloud environments locally remains a challenge,” conveyed Dr. Bandaranayake.
How Sri Lankan Research is Held Back by Limited Computing Power
The lack of large data storage and cloud computing capabilities are key challenges that Sri Lankan researchers are grappling with today. Quite often, research projects, irrespective of their domain, generate vast amounts of data and require robust computing power to process all of it. Yet, many Sri Lankan researchers at universities and institutes alike, need help with affordable access to these resources. Without this assistance, education and research initiatives by Sri Lankan researchers are believed to be at risk of stagnation, preventing them from keeping up with global advancements.
Illustrating the seriousness of the issue, Dr. Bandaranayake explained, “In the past 4 – 5 years alone, more than 70% of the student projects at the Department of Computer Engineering at the University of Peradeniya dealt with HPC, machine learning, and bioinformatics. As LEARN, we have identified that these projects necessitate substantial computing power and storage capacity, often exceeding available resources. Oftentimes, projects also get stalled due to storage limitations, leading to delays in the publication of our students’ research.”
Further elaborating on the ground situation with the University’s computing infrastructure, Dr. Bandaranayake conveyed that a multidisciplinary research team from the University of Peradeniya and a group of researchers from the Department of Wildlife Conservation had undertaken a study to analyse the data behind the unique black leopard subspecies in 2020. At the time, around 15TB of server space had been allocated for this project; which had run out within one week. “This was a project that could have easily been completed in a month or two, but the absence of storage space and the procurement of a new server delayed this project by many more months.”
In addition to this, Dr. Bandaranayake expressed that the limited computing resources in the country foster unhealthy competition among students for the use of limited resources. Consequently, international research projects in prominent domains, such as climate research, which requires petabytes of storage beyond Sri Lanka’s current capacities, are being dismissed, leading to significant missed opportunities as a result of the country’s inadequate academic computing infrastructure. Recognising this shortfall and seeking to rectify it is why LEARN has begun proactively addressing the issue to introduce HPC facilities to empower Sri Lankan researchers.
Learning from Regional Supercomputing Powerhouses
To overcome future challenges the country would face as a result of the lack of supercomputing facilities, LEARN has reached out to its regional counterparts. Recently, around the time of the APAN57 Conference, a delegation from LEARN also visited Thailand’s NSTDA Supercomputer Centre (ThaiSC) and Singapore’s National Supercomputing Centre (NSCC). Through these engagements, the LEARN delegation sought to understand the experiences of their regional counterparts in operating supercomputers and how they overcame the challenges to serve their stakeholders. Through informative sessions conducted by the leaders of ThaiSC, LEARN obtained firsthand knowledge of Thailand’s impressive supercomputing infrastructure, exemplified by systems like the LANTA supercomputer. Likewise, during their visit to the NSCC, the LEARN delegation got a first-hand look at the ASPIRE 2A system.
Navigating the operational challenges of supercomputers
One of the biggest challenges with supercomputing is the heavy initial investment and subsequent maintenance costs involved. Sharing what LEARN gathered from its engagements with ThaiSC, Dr. Bandaranyake stated, “In addition to a substantial initial investment, recurring expenses associated with such computing power shows the importance of ensuring the use of supercomputing aligns with the appropriate usage and national research priorities to obtain the maximum ROI.” This was compared using the case of the LANTA supercomputer, which cost around US$ 100 million to set up, and gets a peak usage of approximately 85% of its maximum capacity. For any such supercomputer, an average of between 70% and 90% of its capacity should be utilised to make sure the centre is not running at a loss and also has a significant contribution to national development. “The average cost of electricity for running ThaiSCis about US$ 1 million a month. While this will depend on local utility charges, when setting up such a facility, we must have use cases that deliver a tangible return on investment.”
Recognising the heavy costs involved, the CEO of Singapore’s NSCC, Prof. Tan Tin Wee, emphasised the importance of first developing use cases that require supercomputing capabilities before investing in such facilities. Modern supercomputers like Singapore’s ASPIRE 2A systems can support a variety of fields running from physics to microbiology to chemistry to AI and more. Building upon this idea, the President of SingAREN, Dr. Francis Lee Bu Sung, recommended LEARN explore setting up a specialised HPC facility focused on AI. Such a specialised approach would lower the barrier for a developing nation to build its HPC infrastructure. Prof. Wee also encouraged the LEARN delegation to collaborate with its international partners to share resources. Currently, Singapore has adopted the latter approach, where projects requiring heavy computing power are used by its native supercomputers. However, for research projects that require lower computing power, the NSCC collaborates with its international HPC partners to utilise their excess capacity to meet the needs of these research projects.
Dealing with the fast-release cycles of technology
Another key issue with supercomputers is that they have a prime shelf-life of 3 – 5 years. This phenomenon is due to the rapid obsolescence of technologies owing to the fast release cycles of hardware companies like NVIDIA. Having recognised this, ThaiSC adjusted its procurement process when building the LANTA system. Dr. Bandaranyake explained, “Apparently, ThaiSC’s procurement process can usually take longer time, due to regulations and compliance standards. During this time, there is room for vast advancement in technology. To overcome this, ThaiSC had planned its procurement such that to obtain the latest technology at the end of its procurement process. This is a clear example of planning that Sri Lanka also needs to adopt when setting up its supercomputing facility.”
However, this doesn’t mean a supercomputer becomes unusable in a short span of 5 years. While ThaiSC uses its flagship LANTA system, its earlier TARA system remains in use by researchers for their projects. However, the issue with older systems is that over time maintenance becomes a costly challenge. Nevertheless, having an older system is increasingly proving to be better than having no system.
During their time in Singapore, the LEARN delegation also met the Global Head of the NVIDIA AI Technology Center, Prof. Simon See. He shared that recent advances in technology, such as the rise of AI if left unchecked, will only deepen existing digital divides. Hence, companies with a disposable inventory of last-generation components of supercomputers, such as GPUs, are looking at donating them to help developing nations expand their computing infrastructure.
The importance of forward-thinking policies
Although Thailand’s supercomputing industry took off much later than others in the region, it has quickly become a leader in the field of supercomputers. This is largely due to supportive policies like the Thai National AI policy (Thai NAIS), government and private entity support, capacity building, and regional partnerships such as the Joint Alliance of Supercomputers (ASC). All of which has collectively ensured its supercomputing capacity is utilised effectively to empower researchers across the region. Similarly, Singapore’s NSCC also operates quite flexibly with limited bureaucracy to ensure it can serve its stakeholders effectively.
Notably, amongst Prof. Wee’s compatriots are two Sri Lankan computing legends – Prof. Gihan Dias from the University of Moratuwa and Prof. V. K. Samaranayake, who is considered the Father of Sri Lankan computing. Commenting on these observations, Dr. Bandaranayake stated, “Singapore and Sri Lanka began their computing journeys around the same time. However, Singapore managed to sprint ahead of us. Similarly, Bangladesh, which initiated its NREN much later, has seen greater progress as a result of cohesive, forward-thinking policies and necessary funding.” Ultimately, the story of our neighbouring nations highlights the importance of proper policies that enable an environment of innovation.
LEARN’s vision of robust computing infrastructure for academia
As Sri Lanka’s NREN, LEARN remains committed to removing barriers and empowering local researchers. In line with this commitment, the organisation’s primary focus is on initiating working groups to develop practical use cases that require HPC facilities. By fostering collaborative efforts, LEARN aims to explore the diverse applications of advanced technologies, beginning with the establishment of these working groups. In the near term, LEARN’s strategy revolves around leveraging its regional partnerships to make supercomputing accessible to Sri Lankan researchers. Looking ahead, LEARN envisions a gradual progression towards enhancing storage, computing, and ultimately, its own HPC capabilities.
This collaborative approach aims to catalyse demand which LEARN aims to meet by working with its regional counterparts like ThaiSC and NSCC. Thereby ensuring the effective allocation of its efforts and investments. LEARN’s endeavours to usher in HPC are poised to be a pivotal moment in Sri Lanka’s technological and academic landscape. Despite the complexities associated with operating a supercomputer, the organisation remains steadfast in its commitment to empowering Sri Lankan researchers by leveraging partnerships with its regional counterparts. As such, LEARN is posted to uplift Sri Lankan academia with its resolute vision towards a future where supercomputers will accelerate academic pursuits and empower local researchers to make meaningful contributions to the global scientific community.