ACM TechNews
SIGMETRICS 2007 Panel at FCRC
Vishal Misra from Columbia Univ (the cricinfo guru) organized a panel at SIGMETRICS on performance measurement and modeling in industrial research labs. The panelists were Cathy Xia (IBM), Albert Greenberg (Microsoft), Shubho Sen (AT&T), Arif Merchant (HP) and me, in the order in which we spoke. The focus was on describing the successes, challenges of performance modeling in respective worlds and in general, on being a researcher in industrial labs.Cathy went first, described the great history of queueing systems at IBM, focused on one example of performance modeling for web services, and did a thorough job of describing the life of a researcher at IBM. Albert did a near-live demo of Microsoft tools and talked about how almost all aspects of CS research were represented and relevant. Shubho did a very nice job of communicating the complexity of managing large IP networks, and presented a concrete challenge: look across network layers and abstract a single measure of happiness for an application (or user). Arif spoke about a tiny slice of work at HP and discussed how traditional disk modeling methods are inaccurate for any particular disk system in large data centers, and how detailed modeling is necessary, but the lesson was not to overfit. Since I went last, I had an easy job. Others had already talked about nearly all computer systems (OS , disk, networking, web) where performance modeling is routinely already done, so I could plunge ahead and point out a brand new direction: the market place of advertisements for sponsored search is a complex system that really needs modeling, measuring (eg across search companies) and understanding.
There were a bunch of questions later from the audience (some button-down shirted students!), the ones you would expect (what it takes to succeed in labs, how do I get a job), few somber ones (do you compromise your research being at labs, who tells you what to work on at labs, what are the challenges to getting your ideas into real products), and one quirky question (how is Google different from other research labs). These questions typically take long answers and the panel went over the time significantly. To summarize the answers were: (1) projects, paper and patents, which I will extend to, projects, products, papers and patents. (2) send CV to people you know in the labs, (3) while research labs help in getting a great research career, it is not clear they are ideal if you had a fields medal, godel prize, sized ambition and the problem to work on, (4) most labs support researchers coming up with project ideas on their own, (5) working with product groups is the challenge, and finally (6) I was surprised by how much was common among different labs as portrayed by the panelists.
posted by metoo
4 Comments:
In my experience, one of the great things about research labs is that you can constructively work on "applied theory", designing algorithms that people need and want to use. In academia, the distance from this seems to increase naturally, as publications instead become the primary measure of value.
I agree with Mike. In addition, getting real datasets is (much) more challenging in academia. That said, I of course value the overall independence one has in academia, but hope that more fruitful models for academy -- real-world collaboration will emerge.
aravind
Let me push this a little more (to a hopefully not-yet-dangerous ground). Is there a compromise one makes to their research for the benefit of working in a research lab that gives easier access to real data and helps one do "applied theory"? This was one of the questions that emerged in the panel discussion.
Muthu --
I can't say I like this phrasing. Sure, when you work for a research lab, there's generally an implicit assumption or explicit requirement that you'll do some work that's in the company's interest, however they determine that. How much of your time and effort is spent on such works depends on you and the company.
But the same goes for faculty positions. There are classes to teach, committees to serve on, funding to get, and so on. It sounds like these graduate students have an idealized picture of academia, and are overly pessimistic in their picture of research labs. In my experience, that's typical -- so I encourage students to try a research lab as an intern to see what it's like!
© Copyright 2008 Information, Inc. This service may be reproduced for internal distribution.