Last week, I attended NSDI for the first time in quite a few years… I only managed to be at the conference for a day-and-a-half, but there was a lot of interesting stuff going on even in just that short time.
For me, it’s always stimulating to attend pure systems conferences like NSDI, given the contrast in research style with my own. For example, there were more than a few papers where somewhere in the implementation, a quite challenging resource allocation problem came up, and the authors just applied a simple heuristic and moved past it without a second thought. For me, I’d be distracted for months trying to figure out optimality guarantees, etc. That’s, of course, a lot of fun to do and sometimes pays off, but it’s always good to see a reminder that often simple heuristics are good enough…
If you only look at four papers, which should they be?
Well, of course, you should start with the best paper award winner:
- Software Dataplane Verification by Mihai Dobrescu and Katerina Argyraki, École Polytechnique Fédérale de Lausanne
The topic of this paper highlights that, despite the fact that NSDI is a true systems conference, there were a definitely a few papers that took a theoretical/rigorous approach to design. (Of course our paper did, but there were others too!)
For me, two of the highlights in this respect were:
- High Throughput Data Center Topology Design, by Ankit Singla, P. Brighten Godfrey, and Alexandra Kolla.
- Building Web Applications on Top of Encrypted Data Using Mylar by Raluca Ada Popa, Emily Stark, Steven Valdez, Jonas Helfer, Nickolai Zeldovich, and Hari Balakrishnan.
These are on completely different topics, but they both really followed a rigorous design approach. The first one was one of the few papers at the conference to actually have a theorem in the body of the paper. (In ours, we left the theorems for the appendix…)
Of course, our paper was a highlight for me too, so I’ll make that the fourth on the list:
- GRASS: Trimming Stragglers in Approximation Analytics by Ganesh Ananthanarayanan, Michael Chien-Chun Hung, Xiaoqi Ren, Ion Stoica, Adam Wierman, Minlan Yu.
The setting for this work was really driven by Ganesh, Minlan, and Ion, but Xiaoqi and I were happy to help with the design. Systems for analytics like Hadoop and Spark are quite difficult to attack with theoretical models, since the complexity of the systems quickly makes any realistic model too complex to provide theoretical guarantees. But, we found that a very simple model still gave useful design guidelines for how to mitigate the impact of stragglers, and at the end of the day, the design came out quite elegantly, at least in my opinion. Perhaps more importantly than that, we were really able to be quite close to optimal in the benchmarks that we looked at.
The operational systems track
Another highlight for me was the operational systems track, which I think is a great idea for conferences in this area. This track included three papers that were focused on real, large-scale, deployed systems.
- Network Virtualization in Multi-tenant Datacenters by Teemu Koponen, Keith Amidon, Peter Balland, Martín Casado, Anupam Chanda, Bryan Fulton, Igor Ganichev, Jesse Gross, Natasha Gude, Paul Ingram, Ethan Jackson, Andrew Lambeth, Romain Lenglet, Shih-Hao Li, Amar Padmanabhan, Justin Pettit, Ben Pfaff, and Rajiv Ramanathan, VMware; Scott Shenker, International Computer Science Institute and the University of California, Berkeley; Alan Shieh, Jeremy Stribling, Pankaj Thakkar, Dan Wendlandt, Alexander Yip, and Ronghua Zhang, VMware
- Operational Experiences with Disk Imaging in a Multi-Tenant Datacenter by Kevin Atkinson, Gary Wong, and Robert Ricci, University of Utah
- VPN Gate: A Volunteer-Organized Public VPN Relay System with Blocking Resistance for Bypassing Government Censorship Firewalls by Daiyuu Nobori and Yasushi Shinjo, University of Tsukuba
The first two are about “classical” NSDI topics — network virtualization (at VMware) and multi-tenant data centers (running Emulab) — but the third was quite different. It talked about the cat-and-mouse game being played to create a VPN relay system for getting around the “great firewall” of China! The talk was quite intriguing, as it went through the timeline for the design changes they’ve done, and how the Chinese firewall responded…
A last note
I should say that (among the talks I saw) there was a clear winner of the “best presentation” award… I’d never seen James Mickens give a talk before, and I must say it was impressive. I’ve never seen someone give a more entertaining conference talk… The NSDI one isn’t freely available, but here’s a link to a talk of his on web browsers. Sounds dry? Well, I guarantee you’ll be entertained if you watch!