Canadian university researchers are a year ahead of schedule in creating a networking test bed on which concepts for building a next-generation Internet can be explored.
The achievement was celebrated last week at a conference at the University of Toronto for academics and corporations including IBM Canada, Cisco Systems Canada and Juniper Networks behind a project called SAVI, short for strategic network for smart applications on virtual infrastructure.
“We have a five-node test bed. The eastern most point is in McGill (University in Montreal), the eastern-most point is in Victoria, and we also have three nodes in Ontario. The ones in Ontario are connected directly by Ethernet, so there’s no IP (Internet Protocol). The others are connected by IP, but we are currently in discussions to have them connected as well via Ethernet.
“Once we have this, we have a facility to conduct research in a way that allows you to validate that the concepts are deployable in real environments.”
Other research groups around the world are also working on next-generation networking, he noted, a number of which have already created well-established test beds and are conducting experiments for a few years.
“SAVI as a very recent test bed has the advantage of addressing the most current challenges — convergence of cloud computing and software-defined networking. So SAVI is positioned to contribute at the leading edge of current issues and challenges.”
Currently SAVI is made up of groups of servers with about 500 processor cores, more than 10 field programmable gate arrays (FGPAs) systems, and about six graphic processor systems all connected to roughly 50 TB of storage. The network is made up of 10/1 Gigabit Ethernet fabrics using the OpenFlow protocol to separate the forwarding path from the routing control. The Gigabit Ethernet network backbone in Ontario is provided by Ontario’s ORION research network and the national IP connectivity is provided by the CANARIE network.
SAVI merges network and compute infrastructures using the OpenFlow protocol for software-defined networking (SDN), and OpenStack for creating a cloud computing platform, to allow new kinds of applications to automatically configure carrier and enterprise networks.
Although it includes SDN, Leon-Garcia said SAVI goes beyond that to create software-defined infrastructure (SDI). One of its successes has been the creation of an SDI manager.
Thomas Lin, a U of T grad student who is on the SDI team, explained in an interview that the manager –simply put, a database of resources — lets users spin up compute, network and software resources needed for an application.
SAVI often gives the example of a wireless network automatically expanding capacity to support tens of thousands of people at a Toronto Blue Jays game sending out videos, stills and tweets, then contracting as the game ends.
But in an interview Friday as the conference ended Leon-Garcia said the network he envisions could do much more — for example, route vehicular traffic in cities and on highways in part from wireless data from cars and trucks.
SAVI quietly went live within the last two months, although it has a few months to go before it is robust enough be used for experiments. It sometimes crashed during demonstrations at the conference.
However, just because the test-bed is built doesn’t mean that SAVI’s approach has been validated. It only means that soon it can be tested.
In pursuit of the goal of an application-dominated network, the SAVI team is split into five areas: creating reusable frameworks for smart applications, designing an adaptive cloud computing platform; designing a smart-edge converged infrastructure to support apps, create a high-bandwidth virtual access network using dense small cells, and the test-bed for testing an entire solution.
SAVI gets the bulk of its funding — $1 million a year over five years — from the Natural sciences and Engineering Council of Canada (NSERC). Almost an equal amount of funding in kind a cash from carriers and network equipment makers (including IBM [NYSE: IBM], Cisco [Nasdaq: CSCO], Juniper Networks, Telus Corp., MTS-Allstream, Ciena Corp., the national Canarie high speed research network Ontario’s Orion research network, Alberta’s Cybera research network, Ericsson’s BelAir Networks division and others) who want have a stake in work that could give them an edge over competitors.
These partners provide researcher time, host students, and donate equipment and software.
Stephen Perelgut, university relations manager for IBM Canada, said in an interview at the conference that his company has paid for some of SAVI’s research as well as donated equipment, said one of the things it hopes to get out of the project is access to talent.
Next-generation networking is a complicated area, and behind able to recognize students who have knowledge about it “is gold,” he said in an interview.
Beyond that, what impressed him at the conference was the number of research papers presented — he estimated there were 30 — on various aspects of networking” there are some really exciting things going on.”
Research is “further along than I expected it to be,” he said.
The use of FGPAs at the edge of a SAVI cluster to accelerate certain tasks “is very exciting, because the next generation of computing is going to have to be very clever about how it runs.” FGPAs are currently used in cellular towers.
Zouheir Mansourati, vice-president for fixed network planning and engineering at Telus Corp. and vice-chair of the SAVI board, said the creation of the test bed lends a lot of credibility to the work of the Canadian researchers.
“We are constantly looking for new ideas,” he said in an interview at the conference in explaining Telus’ support for SAVI. “You have here a wealth of researchers young and old. By having our employees engaged in coaching and having students work with us it gives us a bunch of things – access to ideas, but also access to talent.”
SAVI’s work is of interest to a carrier in a number of areas, he said. These include how problems of security and latency will be addressed, the use of OpenFlow to overcome heterogeneity of network equipment, and how robust SAVI’s network can be.
