Research Spotlight

System Energy Efficiency Lab (UCSD)

I’ve been meaning to write overviews of different research groups who do interesting work in related areas. Fortunately, we happen to collaborate with many such research groups through the TerraSwarm mega-grant. I recently found myself assigned to a project that is primarily directed by folks at UCSD. Figuring out who is who, who does what, and who needs to know about my existence and doesn’t yet has made getting involved more challenging than I was expecting. By this point I think I have figured everything out, but the experience has been good motivation for me to discover in a systematic way just who is over there at UCSD.

The System Energy Efficiency Lab

The System Energy Efficiency Lab (SEELab) is directed by Prof. Tajana Rosing. SEELab explores energy consumption at all scales, from CMOS design to mobile devices to datacenter operations to the power grid.

People

Prof. Tajana Rosing. Tajana is the director of SEELab, and the lead on the SmartCities theme of TerraSwarm. In 2001 she received a PhD from Stanford on “Dynamic Management of Power Consumption.” While getting her PhD she began working at HP, and after she got her degree she continued her work there, while maintaining research engagements at UCSD (only now on behalf of HP). After three years she became faculty at UCSD, and three years after joining UCSD she picked up an additional appointment on the executive board of the San Diego Supercomputing Center. (Her lab definitely has a large-scale systems flavor between that appointment, the datacenter research, and the MuSyC mega-grant she was also a theme leader on.) She also led the NSF CitiSense project, which received press. CitiSense was broadly about designing and deploying a pub/sub architecture called Open Rich Services (OSR) in order to provide infrastructure support for city-scale mobile (phone) sensor data. The goal was to gain insight on the individual impacts of citizen behavior on the environment and health using community-based sensing. She has published over 150 publications, received best paper awards, given invited talks, and managed over $130 million in grant money in total.

Baris Aksanli. Baris is a fourth year PhD student. Holy cow this guy has a lot of publications! Fourteen! A couple are best papers and a paper at ISCA. He was a double-major in computer engineering and math at Bogazici University in Turkey. Now he is doing smart grid stuff, but he also has gotten a lot of mileage (read: papers) out of datacenter efficiency. This is also the guy who seems to be the point of contact for organizing collaboration, since he told me he’s the one I should email if I have future telecon/coordination issues.

Alper Sinan Akyurek. Sinan is also a fourth-year PhD student, has a BSc and MSc already from Middle East Technical University (METU) in Turkey. At METU he was a part of a research group that focuses on the algorithmic side of network communications (especially over wireless links), but his PhD research area is in grid-level power systems operations, specifically smart grid behavior (which I guess makes sense – still looking at network flow dynamics). Also, he worked for five years before the PhD. His up-to-date resume has five publications, four conference papers and one journal article.

Jagannathan (Jug) Venkatesh. Jug is yet another fourth-year PhD student. He is interested in smart grid stuff as well as energy efficient mobile computing. He has five papers including a best paper. (Maybe I should check out this ISCC conference I keep seeing, and the HotPower workshop?) He went to UVA for undergrad (woo!) and double-majored in EE and CS.

Christine Chan. Christine isn’t working on the S2Sim project, but she is here at the TerraSwarm localization workshop which kicks into gear tomorrow, so I figured I should take a more in-depth look. Christine is a third-year PhD student, who got her BS in computer science from UIUC. She has two publications, and so far she seems to be really focused on low-level energy efficiency, like server-level. However, her research interests on her workshop introduction slides says she’s interesting in context-awareness for building automation, so maybe she’s switching scale.

Projects

I looked over the research projects, but basically the bios of the people above seem to capture the high-level picture. SEELab’s current projects are on smart cities, datacenters, multi-processor systems (thermal), and mobile phone platforms.

Summary

Tajana currently has one postdoc and eleven PhD students. The eleven students seem to be fairly evenly divided between a large-scale (datacenter/smart grid) focus, a low-level (processor/machine level) focus, and a mobile/smartphone focus. There might be more towards the large-scale end, but surprisingly there doesn’t seem to be one area that dominates the rest. Interesting!

What I find so impressive about Tajana’s research group is that while the students seem to have strengths in a particular scale/system, in general people in the group seem to be able to move from micro scale to macro scale with no problem. Very flexible! When I first read SEELab’s description I thought that there was no way they could really be going into such depth on energy consumption at all those different levels of scale, and that surely one would dominate, but it turns out that the group is very balanced!

Uncategorized

Flurry: An Affront to Privacy

I was looking at this graphic on the Washington Post about smartphone usage when I came across the following text:

Flurry suggests that “smartphone addicts” – people who open apps 60 times or more a day – are growing at more than five times the rate of regular users. In fact, while Flurry reports that its regular users (people who open 16 apps a day or less) have increased 23 percent, to 784 million people, super-addicts have grown 123 percent – to 176 million.

Flurry gathered that data from the more than 1.3 billion devices it tracks for customers like Pinterest, Snapchat and Zynga.

Wait, what?

That’s right, you’ve probably never heard of them, but Flurry Analytics tracks more smartphone users than Google or Facebook.

Did you know that if you happen to have one of the over 500,000 apps (including Shazaam and AccuWeather) that use Flurry for their analytics (Forbes estimates that your phone has 7-10 of them), then Flurry gets all of your screen views and clicks within the app? It also collects the usual suspects like your phone’s OS version, unique phone identifiers, your location, and so on. You might think that maybe this is okay because you have agreed to use Shazaam’s services, so Shazaam should be able to get usage data via whatever middleman it wants. However, from a single app Flurry also appears to track when you launch and use any app on your device, even those that do not use Flurry, such as Facebook and native browsers (Android allows apps to access real-time information about other apps installed or running on your phone. I don’t know about iPhone/Windows/Blackberry). Flurry shares aggregate information about your behavior across all apps with all its customers, not just the ones whose apps you agreed to install.

Not only that, but based on your usage of the suite of Flurry apps installed on your phone and at least metadata about the others, Flurry can classify you under several different Personas. Are you a Business Traveler? A New Mother? How about LGBTQ? (Note 9/15/14: At this time they have removed LGBTQ from their webpage as an example Persona.)

Did you know that Flurry auctions your screen activities to advertisers in real-time?

If your lucrative demographic is no longer spending time and money in a customer’s app, Flurry can bring you back into the fold with pop-ups in other apps.

As a researcher, I appreciate that there are many insights that you can only get from a large amount of global usage data and analytics. But as a citizen, and from an ethical perspective, I find this creepy as hell.

As much as I would like to point my finger at Flurry, the real problem is a systemic one. There is a level of indirection between end users and analytics services that completely obfuscates the who, what, when, why, and how of data collection. Flurry’s privacy policy places all responsibility for disclosure and for providing the Flurry opt-out link on the app developers, though it could easily do those things itself. App developers have zero to negative incentive to expose an interface to Flurry, themselves often willfully ignorant of the kind of data Flurry collects in order to provide its services. The lack of transparency about pervasive tracking both on mobile devices and the web comes from a perversion of computer science’s most useful tool: abstraction. I believe computer scientists have an ethical obligation to recognize, fight against, and help correct the structural roots of these kinds of unhealthy mass trends, but unfortunately many of our best and brightest have dedicated themselves to jobs founded upon increasing ad-based revenue, and having bought into the hype, turned a blind eye to the questionable ethics of their industry. Let’s hope the bubble bursts soon.

Here is what Flurry does: http://www.flurry.com/solutions/analytics

Here is more of what Flurry does: http://www.flurry.com/legal-privacy/privacy-policy

Here is the opt-out page: https://dev.flurry.com/secure/optOut.do

If you are an Android user, you need to provide your Android Device ID to opt-out. To get your Device ID, it appears you have to install and app like Device ID by redphx (it seems okay, no permissions needed).

Computer Science

Conference Spotlight: NSDI’12

Name: The 9th USENIX Symposium on Networked Systems Design and Implementation (proceedings)

Date: April 25 – 27, 2012

Location: San Jose, CA

Overview: Mostly three (sometimes two) papers per session. Eleven sessions, so less than 33 papers. Two sessions on Big Data (second session actually parallel computing!), two on security/privacy, two on datacenters/cloud. Relevant sessions: one on wireless (maybe too low-level), one on new architectures and platforms.

Authors: MSR and UC Berkeley seem to be competing for domination of this conference. Other top industry and universities are present. Surprising number of Wisconsin papers, but it turns out that they’re ranked about the same as Michigan (which is of course awesome). I should pay more attention to Wisconsin.

Relevant/Notable Papers:

Best Paper: “Resilient Distributed Datasets: A Fault-Tolerant Abstraction for In-Memory Cluster Computing“. UC Berkeley, including Ion Stoica who worked on Chord, which is very cool. After reading the abstract for this one, I’m not sure why it’s so cool. Will have to read/watch talk to get it.

Posters/Demos: Only thirteen! Notable projects: Carat, UW’s third party tracking in the wild, PhoneLab. A couple of anomaly-detection things, including anomaly-detection in sensor networks.

Notes: Serval seems useful for sensing – splitting the service-level control and data planes to better serve mobile or multi-homed clients. But then there’s XIA, which says that switching to another newfangled internet architecture still means the losing architectures can’t play. XIA was designed to support interoperability between all of them (should read, sounds good).

Multi-path TCP paper. Might be useful for I/O over different channels. Maybe useful for software-defined lighting? Downlink through light, uplink through radio? Downlink would need to be infrequent or low bandwidth, such as control sequences or commands for a phone. (Broadcast data like room/light ID doesn’t require uplink.) The multi-path TCP paper, like so many others, cited the characteristics of mobile devices, datacenters in its motivation. These are apparently what we care about these days.

What is content-aware networking? What is distributed differential privacy?

At first, I didn’t think that NSDI was going to have much to offer as far as wireless sensor networking/smart environments go. However, networking in general deals with many basic concerns and themes that are pretty relevant to sensor networks. There is accounting and attribution, malicious subsets of the network, unforeseen global behavior arising from decentralized interactions, generalization and abstraction for handling heterogeneity, dealing with churn, addressing resources, privacy protections for location data, and much more. Very interesting conference.