UPDATE: This article has been updated to clarify which institute is responsible for the Anton program and to clarify earlier comments about Anton 1's fate.
NEW YORK (GenomeWeb) – The National Institutes of Health has renewed its support for a program run by the Pittsburgh Supercomputing Center (PSC) that provides free access to a special purpose supercomputer called Anton, developed by D.E. Shaw Research, which is designed for molecular dynamics simulations of biomolecular systems.
The NIH has awarded a $1.8 million grant to PSC to make a second iteration of the supercomputer, Anton 2, available to the research community for non-commercial use starting in the fall of 2016. The 128-node system that will be installed at PSC later this year will expand on the power and capabilities of its predecessor — the first Anton, which is currently in operation at PSC — including increasing simulation speeds by approximately fourfold and enabling the simulation of biomolecular systems with around five times as many atoms. Full details of the Anton 2 system are described in a paper published in the 2014 Proceedings of the International Conference for High Performance Computing, Networking, Storage, and Analysis.
Philip Blood, principal investigator of the new grant and senior computational scientist at PSC, told GenomeWeb that the NIH grant will fund the PSC staff who will be responsible for the day-to-day running of the system and for supporting the researchers that are using the system to run their simulations. The funds will also cover the costs of file systems and storage servers and other infrastructure needed to support projects as well as to cover power and cooling costs. PSC will also organize workshops that introduce new users to the system's specialized architecture and hardware. The funds will support Anton 2 through the year 2020.
The new grant will also extend Anton operations until the new system is in place. Anton has been in use since the fall of 2010 at PSC, where it has supported researchers' efforts to better understand the behavior of proteins, cell membranes, nucleic acids, and other molecules at the atomic scale. The NIH's National Institute of General Medical Sciences awarded a $2.7 million grant to PSC that year to foot the costs of hosting, maintaining, and running Anton. Following a successful first round, the program was extended in 2011. So far the supercomputer has been used in 277 simulation projects conducted by 127 different principal investigators.
The new system is similar to its predecessor in that it uses a network of custom application-specific integrated circuits to perform the simulations. However, Anton 2 offers architectural improvements including a new network structure and more computational units than its predecessor. It features more central processing unit cores and more flexible fixed-function pipelines and also offers a higher peak throughput and CPU clock frequency than the first iteration of the system. It also offers greater capacity per ASIC and a higher communication bandwidth between ASICs.
"On Anton, people could simulate systems consisting of up to about 150,000, maybe 200,000 atoms," Blood said. "On this new system, we are anticipating that the size of the [simulated] systems will be roughly 5 times larger ... around 700,000 atoms depend[ing] on the type of system." These updates should make it possible for researchers to explore a much wider range of biological systems, for example, complex multicomponent biological systems. "There are a lot of possibilities for people to get very creative and to be able to simulate biological systems that were not possible before," he said.
Cloud infrastructure has become popular in the life science research space because it offers relatively cheap and easy access to hardware on a large-scale. But supercomputers like Anton have remained a mainstay for molecular dynamics researchers. "Molecular dynamics groups have a long history of effectively using the national supercomputing centers —as well as campus infrastructure — to run their simulations," Blood said in an email. "These resources are by far the most popular for running MD codes, which have been highly optimized for these systems."
Part of the reason for this is that cloud technology is ill-suited for the molecular dynamics space. "Most cloud computing resources are better suited to uncoupled or loosely coupled problems, since network performance between nodes is not usually great," Blood explained. "Molecular dynamics is a tightly coupled problem, and so does not perform very well on most cloud computing infrastructure. You can buy time on infrastructure that supports tightly coupled problems, but it is much more expensive."
An exception to that are "problems that require many small MD calculations, with each calculation contained on a separate node," he said. "This type of work would be amenable to cloud computing, but people have also used free resources like Open Science Grid and even the computers of regular people who agree to let MD codes run on their computers when they are otherwise idle, [for example,] the Folding at Home project."
PSC will continue to allot time on Anton 2 in the same way it allotted time on the first Anton. "One of the things that's important to us is that as many research groups as possible as want to use Anton, who submit competitive proposals, can get access," Blood told GenomeWeb. "The goal is ... to have a balance between the breakthrough timescales but also allow many different research groups to benefit."
In terms of projects, Blood expects the system to support a number of protein-folding projects, but also efforts to simulate signal transduction mechanisms as well as those of ion channels and transporters in cells. "These have all been big targets on the existing Anton system. We expect more of that ... with even bigger systems and longer timescales," he said.
As has been the case in the past, interested researchers will be required to submit research proposals first to PSC, where they will be pre-assessed to ensure that they are a good fit for Anton. Submissions that pass muster will then be passed on to an independent expert committee convened by the National Research Council at the National Academy of Sciences that will make the final selections. Typically, PSC publishes a request for proposals in the spring with applications due in June. "The timetables may not be exactly the same this year but I think we'll try to keep them close to the same allocation cycle as much as possible," Blood said.
In the intervening months, the PSC team will work on installing Anton 2 and prepping for the first set of allocations in the fall. It's not clear what will happen to Anton 1 when it goes out of commission and is returned to DESRES. "That's up to D.E. Shaw research," Blood said. "They still own the machine."