U.S. research and education (R&E) network Internet2 is working with other global R&E networks to enhance resiliency, bandwidth and economics of undersea connectivity. As Internet2 President and CEO Dave Lambert and Rene Buch, CEO of NORDUnet, explained in an interview, R&E networks also may be collaborating on cutting edge projects involving virtual supercomputing. NORDUnet is the R&E network for five north European countries, including Finland, Sweden, Norway, Iceland and Denmark.
Global R&E Networks
The global R&E networking community realized a few years ago that, collectively, they had purchased 26 separate connections across the north Atlantic that were “paid for by separate parties with no coordination or backup arrangements,” said Lambert.
There were two main things that were big issues, noted Buch. None of the 26 logical links were purchased together, which means there were “no economies of scale and no redundancy,” he said. Some of the connections were on the same physical link, which meant that if that link were to go down, all those connections would be lost. Similar issues were a concern for other R&E network undersea connections as well.
Various global R&E networks decided that it was time to “move in stages to a more coherent and rationalized infrastructure,” said Lambert.
They started by working on defining a Global Network Architecture (GNA) – the first public version of which was released publicly earlier this year. As Buch explained, the architecture addresses technical issues such as how to report errors from one network operator to another and how to know whom to call to get issues resolved.
One of the first groups to comply with the GNA is the Advanced North Atlantic (ANA) collaboration, which consists of Internet2, NORDUnet and four other R&E networks – CANARIE (Canada), ESnet (USA), GEANT (Europe), and SURFnet (The Netherlands).
ANA members now share a total of 740 Gbps across the Atlantic. Each participant is guaranteed a certain amount of bandwidth. The rest is “donated to a common pool,” explained Buch.
Participants can burst above their guaranteed data rate when bandwidth is available – an approach that makes sense, considering that participants operate in different zones with different peak times.
Recently, the NEAAR Project, coordinated by Indiana University and funded by the US National Science Foundation, turned up a new 100 Gbps circuit on a brand new trans-Atlantic facility, which enables ANA members to have backup capability with geographic diversity.
The upshot is that participants now get “more capacity and more resiliency at a lower price,” Buch said.
Virtual Supercomputing
It’s always interesting to talk to people like Lambert and Buch because R&E networks often pioneer new technologies that eventually make their way onto commercial networks.
I asked Lambert and Buch what cutting edge technologies the R&E networking community is exploring at this time and they mentioned virtual supercomputing and SDN.
The idea of virtual supercomputing is driven by the same sort of cooperative spirit that drove the GNA. The goal, Buch explained, would be to put computational capability and storage at open exchange interconnection points that could be combined to create the virtual supercomputer.
The virtual supercomputer currently isn’t a formal project but as Buch explained, “it’s something we’re talking about for the future.”
Another area under exploration would be to build a new layer of services into the network designed to move a data flow pertinent to a specific science, Lambert said. He pointed to the example of output from a genomic sequencer feeding into a particular database or from a telescope into a distributed database structure.
Considerable innovation would be necessary to achieve this, Lambert noted. “It would require us to be much better, not just in how we work across the global layer but also the national layer and campus layer,” he said.
Software defined infrastructure from the edge to the core of the network would likely be required to support end-to-end performance, “particularly when it pertains to characteristics of a particular data flow,” Lambert said.
Earlier this month, Internet2 announced that Howard Pfeffer will be the organization’s new CEO effective June 12.
This is an example of a good idea. We have been developing almost the same concept for the last three and half years. It's called InfiniCortex. Please check the papers:
1.Marek T. Michalewicz, et al. , „InfiniCortex: concurrent supercomputing across the globe utilising trans-continental InfiniBand and Galaxy of Supercomputers”, Conference Paper · Conference: Supercomputing 2014: The International Conference for High Performance Computing, Networking, Storage and Analysis, At New Orleans, LA, USA November 2014, DOI: 10.13140/2.1.3267.7444
2.Marek T. Michalewicz, et al., „InfiniCortex: Present and Future”, CF '16 Proceedings of the ACM International Conference on Computing Frontiers, Pages 267-273, Como, Italy — May 16 – 19, 2016, ACM New York, NY, USA ISBN: 978-1-4503-4128-8 doi>10.1145/2903150.2912887
3. 4.Gabriel Noaje, et al. “InfiniCortex – From Proof-of-Concept to Production”, Supercomputing Frontiers and Innovations, 4 xx-xx (2017) (in press)
4. Jakub Chrzeszczyk et al. "InfiniCloud 2.0: distributing High Performance Computing across continents", Supercomputing Frontiers and Innovations, 3 no. 2 (2016) DOI: 10.14529/jsfi160204