Where Have All The Systems Problems Gone?

May 21, 2007 8:12 AM

For CompSci Researchers: Where do you get your ideas/inspiration from?

I'm a compsci researcher in a UK institution, specifically doing systems research (more specifically, OS & network research).

The UK model for research is quite different from the American model. Here it is acceptable to pick a topic, ask a question and plug away at it eventually finding an answer -- even if it's totally obtuse and obscure. Hence a lot of the research is good, but very niche... -- and doesn't get published apart from tech reports and internal documents.

I find the American model of pushing-the-envelope very satisfying. Reading the proceedings of every SOSP/OSDI/USENIX makes me reel with wonder at the beautiful and elegant solutions to problems-I-didnt-even-know-existed. I am at the point now where I itch for really good problems in the systems field but can't find any for the life of me. Note that I am in a group where people are doing systems research but these tend to be in the form of yet-another-distributed-file system or yet-another-scheduler etc... nothing that is topical and relevant, or even zany. And our abysmal publishing record proves it.

So my question to all you American CompScis, and systems compscis in particular is -- where do you get the inspiration for your topical and timely research? Is it professors? industry? How could a person sitting in the UK come into contact with these inspirations?

Any help appreciated. Thanks.

posted by gadha to Education (10 answers total)

I'm not in CompSci but I am in research in the UK. I've always found the way to get ideas is to read around your subject a lot then spend some time thinking about stuff. (Which I guess could be interpreted as: Get your inspiration by reading journals, etc, where stuff is published in your area of interest.)
Look not just at outputs but also at methodologies.
Think about how things work and how they might work better.
Pay attention to topics not apparently directly related to your core area, you never know when you might be able to apply one thing elsewhere.
posted by biffa at 8:53 AM on May 21, 2007

Response by poster: excellent point biffa, but that's where I'm failing -- "thinking about how things work". If somebody comes out to me and points out a problem or asks for a solution to a problem I can deliver without problems. It's finding the juicy problems that seems to elude me. Usually I just go "ok, this happens like X, if I did it Y would it be of any benefit? yes.. ok let's do it.." which is a poor way of picking research topics of impact
posted by gadha at 9:19 AM on May 21, 2007

Some of the best research has been kicked off by re-examinations of classically intractable problems, such as the four color theorem, which generated the first "computer proof," or the ellipsoidal (1979, Leonid Khaciyan) and interior point (1984, Narendra Karmarkar) methods of linear programming solutions to feasibility and resource allocation problems. So you might look to catalogs of intractable problems for inspiration. But that is, admittedly, a hard wall to check for chinks, and a so, an often discouraging one. But, sometimes, bouncing off the intractable is the best way of finding research traction.

Another source of inspiration might be to talk to managers and executive users of commercial operations research, who are often on the leading edge of systems research with respect to demands of the user base, but who traditionally have fairly conservative risk profiles, because of the need to deliver short term profitability in order to continue funding for further projects. Often, because of the fact that they percieve problems they are asked to deal with as classically intractable, they avoid tackling the biggest problems of a business, in favor of doing work that they can more reliably predict results and benefits of doing. Yet they are often happy to suggest higher risk problems, which they tend to feel are more "basic research" problems , than commercialization opportunities. What are you doing to develop and then mine a corporate OR network?
posted by paulsc at 10:09 AM on May 21, 2007

The professor who taught my operating systems class this year specializes in security, but much of his work has an interesting twist, in that it is inspired by the defense mechanisms present in biological systems, specifically immune systems. He's a nice guy, maybe you could ask him.

I bet you could get some interesting ideas in a similar way. By reading about something outside your field you maybe you'll find some interesting type of behavior that you can use to solve a problem in the systems field, or maybe you'll think of a way to use your knowledge of systems to solve an interesting problem in some other field.

The field you take your inspiration from probably doesn't even have to be as far from computer science as biology. I bet you could find interesting ways to use cellular automata, or graph theory, or neural networks, or any number of similar cool things, to solve systems problems.

Are there any systems problems you know of that are considered to be unsolvable? Attack them with crazy (biological, graph theoretical, etc) ideas and see what you get!
posted by benign at 10:12 AM on May 21, 2007

The thinking about step is dependent on you being up to date on steps 1 & 2. Have you already done that? Read up, identify a problem/hole within the current research, then start the thinking. Once you get to the edge of current research you will increasingly find yourself saying 'I can't seem to find anything about X' or 'This has been done this way but wouldn't in to make sense to do Y instead?' Prior to that you may find times when you read up, think of something but then look up something and find someone has done the work already. The more you read, the more holes in the literature you will find.

I'd also add that you should talk to people in your field regularly. Research in a vacuum is much more difficult.
posted by biffa at 10:15 AM on May 21, 2007

It sounds like you're looking for something a little less applied, but the biological sciences are full of interesting computational questions right now.
posted by chrisamiller at 11:29 AM on May 21, 2007

Response by poster: no I'm looking for VERY applied problems. Problems that are real-world, at least thats what gives me a buzz when I do systems research.

__or__

problems that are uniquely clever and different...
posted by gadha at 11:40 AM on May 21, 2007

I second benign's idea; think in an inter-disciplinary fashion. The stuff I do is at the intersection of CS and biology, and it's amazing how often we come across the same basic idea across both domains, only expressed in each in different language (eg. Adleman came up with the idea for DNA computing by reading about polymerase, the enzyme that replicates DNA inside the cell. It was then but a short leap to the operation of a Turing Machine...)
posted by gene_machine at 1:38 PM on May 21, 2007

The mighty Google knows all.
posted by chairface at 3:45 PM on May 21, 2007

My current professor (who does AI/learning research) says that the way to success is to read a lot of papers, until you find one that is interesting, but flawed in some significant way. Then fix the flaw.
posted by crinklebat at 9:15 PM on May 21, 2007