The University of Auckland is one of ISIF Asia’s two 2022 award winners, for research and early implementation work on Network Coded Tunnels for Satellite Links. This research focuses on improving the quality of connections that use satellite links. Project Lead Dr Ulrich Speidel describes the background behind the research below.
In research, most journeys start on the shoulders of giants. Our project on network-coded tunnels for satellite links is no exception. Before ‘Etuate Cocker and I got involved, there had been many years of network coding research. I had sat through numerous conference presentations on the topic by notable information theorists, seeing the theoretical beauty but not quite how this could be turned into something practical. Applications talked about typically involved some hypothetical protocol for providing wireless HD video to large crowds in a football stadium. Sure, challenging, but nothing that spoke to me — maybe I’m just not enough into spectator sports?
Instead, we looked at the arrival timing of packets that had been traversing the Internet. Call it jitter if you wish. At the time, an almost unnoticed cottage industry had sprung up in New Zealand — call centres that used Voice over IP to provide services to callers in the UK, Ireland, or North America, for example. We wondered how sustainable this would be, given rampant Internet growth and the fact that on-time packet delivery hadn’t ever been an explicit design goal. Everyone knows how a ‘bad line’ slows things down considerably and eventually makes a call centre uneconomical.
So, having a topic, we set up ‘beacons’ all over the globe to measure how badly packet streams would get distorted en route. ‘Etuate seemed to know a boffin on almost every island around the Pacific, which gave us a nice little network in this region. Running the project was a challenge in its own right — beacons failed, got disconnected, hosts retired, universities blocked access under new policies, and power outages took their toll. Backend servers collecting the data also played up. But in the end, one of the big takeaways was that satellite-connected islands were essentially hostile territory for anything real-time Internet.
Back to the giants. Standing on a giant’s shoulder can give you an angle from which to view things differently. Our giant #1 was Muriel Médard from MIT, whose keynote speech at a conference in Haikou about combining network coding and TCP in her lab I had the fortune to attend. Shortly afterwards, I was shoulder-tapped to organize a series of Vice Chancellor’s lectures at the University of Auckland — and Muriel was an obvious speaker to invite. This gave us an opportunity to talk about our island woes. Having spent time on islands as a teen, she got on board immediately, and ‘Etuate and I ended up with an idea of how we could make this work.
Moreover, Muriel and her German colleague Frank Fitzek, teaching in Denmark at the time and giant #2, had a startup in Aalborg called Steinwurf. They had the building blocks of code that we needed. Now all we needed was funding to try this out. Again, giants lent shoulders (I’ll stop numbering them now): ISIF Asia provided funding through PICISOC, as did InternetNZ, not least because several other giants were willing to vouch for us. Many of them are well known in APNIC circles, such as Maureen Hilyard and Brian Carpenter, but also others in Europe and North America.
We got Steinwurf to write some coded tunnel software for us — quite a bit different from their ‘day job’ football stadium-type network coding, but they got on board very enthusiastically.
It also allowed us to head out to four islands, Tuvalu, Rarotonga, Niue and Aitutaki, where the local ISPs and their amazing people (more giants!) helped us set up coded tunnel endpoints and small networks for trial purposes. We got to learn a lot, from finding out how to fly uninterruptible power supplies to islands, to how to bail out someone in Tuvalu when a hotel booking didn’t work out, to figuring out what really bugged the satellite connections in the islands.
The potential hit home when I realized in Rarotonga that I’d left my memory stick with 770 MB of Ubuntu behind and could download a copy in under two minutes via our tunnel across the O3b link — during peak time. My local gentle giant Andrew watched the link utilization during the download and just went ‘whoa’ — he’d never seen it go that high.
Péter Vingelmann from Steinwurf accompanied us online during our deployments in the islands — another giant whose quick response helped solve many problems.
The deployments showed that coding across the links generally led to much better goodput. Transfers succeeded where they would have previously timed out. We learned that no link was the same — some were badly overloaded, some were suffering from TCP queue oscillation, and one, Aitutaki, was awaiting a local network upgrade to catch up with the capacity suddenly on offer.
But we’d kind of cheated. We’d only coded a small part of the traffic on the satellite links but really needed to code all of it. But how would we do that? We only had access to one end of each link and didn’t want to have coconuts thrown at us for causing disruption while we tried things out.
So back to the lab it was. Once again ISIF Asia and InternetNZ came to the party, funding equipment and a partial PhD scholarship to build a hardware-based simulator for ‘a Rarotonga link in a rack’, complete with a ‘world’ to supply data for download and an ‘island’ with hundreds of ‘users’ generating demand.
Our tunnels use some clever maths to send ‘spare parts’ packets that allow recovery of packets lost en route, typically to queue overflows at the satellite uplink modem. As it turned out, this works well if the spare parts get to mix with a lot of other traffic and arrive at the link after it has decongested. But when one codes everything, these spare parts hit the queue while it is still overflowing, and are lost as well.
After experimenting with all sorts of delay schemes that didn’t work well, we realized that spare parts were better sent when the link went idle, but the Steinwurf encoder couldn’t know when that would be. So, with further help from ISIF Asia, we built a ‘titrator’ that works in series with the coder and feeds the satellite link exactly what is needed. No more spare parts displacing other data or getting lost for arriving at the wrong time.
Just as we were looking for a small non-monopoly island ISP to test this, along came COVID-19, and the Pacific Islands and the rest of the world were out of bounds. Staying in New Zealand seemed the only option. We stumbled across an InternetNZ-funded project for a satellite-connected community hub at Te One School on Chatham Island, to be built by Gravity Internet, an ISP run by yet more giants conveniently based one suburb over. Tim Johnson and his colleagues came on board in an instant, as did Te One School and, once again, ISIF Asia and Steinwurf. They’ve all stuck with us through the numerous hoops so far. NZ’s Ministry of Foreign Affairs and Trade (MFAT) also helped with a PhD scholarship for another ICT giant of the Pacific, Wayne Reiher, who finally joined us.
So with COVID-19 now on the wane (hopefully for good), we’re finally getting into gear for implementation on a few real links.
But projects also do more than just technical work — they lead to connections that create new ideas and opportunities for students and other parties involved. For example, going to Singapore to receive an ISIF Asia award at APNIC 54 led to a chance participation in a session on IPv6 deployment. This, in turn, led to a draft inventory of surplus IPv4 address blocks in New Zealand’s public sector and the discovery of millions of dollars of unused IPv4 address space, some of which will no doubt provide a windfall to the New Zealand taxpayer in the coming months.
Giants out there — you rock!
Dr Ulrich Speidel is a senior lecturer in Computer Science at the University of Auckland with research interests in fundamental and applied problems in data communications, information theory, signal processing and information measurement. The APNIC Foundation supported his research through its ISIF Asia grants.
The views expressed by the authors of this blog are their own and do not necessarily reflect the views of APNIC. Please note a Code of Conduct applies to this blog.