This has been my third conference event since I became a PhD student at SANDLab. NSDI 15' was between May 4 - 6, held in Oakland, CA. My friends were saying things like "do not go out at night" because of so many negative news about Oakland downtown. Well, I stuck to what they suggested and stayed in hotel at nights - not only because of the potential dangers but also due to homework due. Yep, skipping two classes and TA discussion is never fun and you have to do the work at night. 😛
NSDI is a conference that pays much attention in practical system designs. Most topics are in data centers, software-defined networks, and operational systems. Although there were wireless and physical layer sections, based on what I observed, most people were more interested in previous topics I mentioned. This year there are three best paper awards, which are by Matthew Grosvenor et al., Dan Ports et al., and Ben Pfaff et al. Here are some papers I think very interesting and I learn new things from.
Queues Don't Matter When You Can JUMP Them!
This is Matthew et al.'s paper from University of Cambridge. Their webpage about QJump system can be found here. The main idea is to 1) limit the rate of inputing packets into the network such that only short queues can exist; and 2) prioritize the traffic in the network, i.e. jump over the queue. If some applications are highly sensitive to the latency, then the traffic from these applications have high priority and can "jump" over the traffic from some applications that are insensitive to the latency. The challenges come from 1) the implementation itself; and 2) the constraints from current data centers, e.g. unmodified kernel, unmodified application, etc. Technical details can be found in the paper. The presentation of this paper is very well organized. Basic logics include:
- Motivation: what is the queuing problem and where does it come from? Network interference and thus delays
- Challenge: where are the constraints?
- Key idea: rate-limiting and queue jumping
- Limitations and solutions
Beyond Sensing: Multi-GHz Realtime Spectrum Analytics
This is a paper by Lixin et al. from MIT. The basic idea in this paper is using USRP to sense wide band spectrum based on the pattern of spectrum usage. The key idea in this paper is that spectrum usage has certain patterns and thus sensing can be scheduled to improve the efficiency. The basic procedure contains two parts: learning patterns and scheduled sensing (predicting). They identify three type of spectrum usages intuitively: always-on, fixed-cycle, and dynamic.
Towards Wifi Mobility without Fast Handover
This is by Andrei et al. from University Politehnica of Bucharest. They suggest a hardware improvement rather than Fast Handover to allow WiFi mobility. From MPTCP, they propose a change such that all WiFi clients should connect to several APs and therefore no need the Fast Handover process. They argue that "mobile client should connect to all APs at any given time." The decision where the traffic should go through is handled by MPTCP. The idea is very interesting and I am more curious about how that would influence the energy, and a hardware change for all could be really tough.
Ripple: Communicating through Physical Vibration
By Nirupam Roy et al. from UIUC. As what they described, the idea is cute but not yet has a practical application that right now needs such idea. Well, who knows? Maybe in the future vibration becomes critical, you never know. I remember a paper from MIT talking about vibration on a chip bag could reveal your conversation, if the spy has a high-speed camera. This paper sort of has the similar idea but extends to other scenarios.
Multi-Person Localization via RF Body Reflections
This is by Fadel from MIT. The techniques are not new and in fact these techniques have been explore in ECE side. The real contribution in this paper is the actual system implementation, and what is really interesting to see is the demo during the presentation. The demo shows the capability of their device to detect the heart beating using a FMCW-enabled RX-TX-RX pair. Although not very accurate, they show the potential that you can use specialized hardware to "spy" on people indoor. Some concerns are mentioned during the conference. The answer to privacy concerns would be not trying to identify a person. I've seen how this could go wrong.. Anyway, his presentation is also very clear.