The tenuous ties between Sigmetrics and CS theory

While I publish in many venues, ACM Sigmetrics is certainly the closest thing I have to a home conference.  It is one of the few venues that can look at a mathematical paper and evaluate it both on its practical relevance and its technical depth — weighing the two together.  As a result, every conference ends up with a mixture of pure systems papers, pure theory papers, and a large number of hybrids with some degree of both systems and theory work.  Of course the balance between these types varies over the years, but in the decade I’ve been involved with the community, all types have maintained visibility.

Given that description, one might expect significant overlap of Sigmetrics with both more systemsy conferences like Sigcomm and NSDI, and more theoretical conferences like STOC and FOCS.  But, while this is true with NSDI and Sigcomm, there is virtually no overlap with STOC and FOCS.  One explanation for this could be that Sigmetrics isn’t really in the middle of theory and systems, it’s biased toward the systems side.  But, I don’t really think that’s true.  If anything, right now it’s biased towards the theoretical side of things.  To me, it seems to be more due to the fact that Sigmetrics uses a different sort of theory than the STOC and FOCS communities — there is much more applied probability and continuous math, because of a focus on stochastic performance bounds rather than worst-case guarantees.  (In the ’80s and early ’90s the theory side of Sigmetrics was dominated by queueing theory, which may be the historic reason for this difference.)

A shrinking divide?

Interestingly, it seems that the gap between the traditional theory CS theory community of STOC/FOCS and Sigmetrics has been shrinking in recent years due to the emergence of areas like “Algorithmic Game Theory” and “Network Economics.”  Questions at the intersection of CS, Networks, and Economics are drawing people from both the theory and the systems communities…  and this time, the techniques used by people across the communities overlap quite a bit.

This really struck me when Sigmetrics handed out its “Rising Star” award this year (an award to the top young researcher in the field).  It went to Augustin Chaintreau, who “grew up” in the Sigmetrics & Networking communities, but these days also has a foot in the EC/WINE community too, which spun out of the FOCS/STOC community as venues for work focusing on Algorithmic Game Theory.  Then, when I looked back, I realized that nearly all of the recent Sigmetrics rising star awardees split time between the EC and Sigmetrics communities.  Going backwards in time, the winners have been

2013: Augustin Chaintreau (press release)
2012: Marc Lelarge (press release)
2011: Adam Wierman (press release)
2010: Milan Vojnovic (press release)
2009: Alexandre Proutiere (press release)
2008: Devavrat Shah (press release)

In every one of these cases, the person does some work in the traditional networking and systems communities (on the theory side), but also does some work in “Network Economics.”  As this list indicates, some of the most exciting work in recent years at Sigmetrics has involved some overlap with economics.

Solidifying the connection

Unfortunately, despite the fact that both the theory community and Sigmetrics have a growing interest in the intersection of CS and Economics, the connection between the communities is still tenuous at best — to the point that in 2013, the conferences were held the exact same week in different locations.  (Luckily, they were in Pittsburgh and Philadelphia so it was possible to bounce back and forth, as many of us did.)

I really hope that going forward, the interaction between these communities grows and solidifies so that we don’t end up with a division in both people and techniques,  as currently exists between STOC/FOCS and Sigmetrics…  A promising step in this direction is that the NetEcon workshop (organized by David Parkes, Jean Walrand, and Patrick Loiseau) was co-located with Sigmetrics this year, and will be again in 2014.  This brought a number of folks from the STOC/FOCS community over to Sigmetrics… hopefully they’ll come back!  Another promising opportunity is that FCRC is coming up, so the conferences will be co-located…hopefully, we can take advantage of that to have some event (maybe a workshop) that brings the communities together.  If anyone has any ideas, let me know!

Advertisements

4 thoughts on “The tenuous ties between Sigmetrics and CS theory

  1. Hi Adam,

    I’m very interested in research that’s somewhere between theory and systems, although naturally I’m more interested in things on the theoretical side. For me personally, I’ve found when doing research that I constantly need to read STOC/FOCS/SODA papers, and I pretty frequently end up reading papers from INFOCOM, SIGCOMM, and NSDI. But I essentially never read SIGMETRICS papers — I hardly ever even see them cited.

    I’m not totally sure why this is, but my guess is that it’s related to what you mentioned — the theory that’s usually at SIGMETRICS tends to be either related to queuing theory or (recently) AGT. I’ve been working a lot lately on network design problems with some practical people, and all of our theory is coming from approximation algorithms and related parts of STOC/FOCS theory. It’s not clear to me if this kind of theory is really appreciated at SIGMETRICS. So we’re deciding whether we should submit to STOC/FOCS (and take out the practical stuff) or submit to NSDI/SIGCOMM (and take out most of the theory). It’s a frustrating place to be in, and I would love it if SIGMETRICS could live somewhere in the middle. But moving towards EC/WINE doesn’t really help me :).

    Love the blog so far!

    • I’d love to see more network approx algorithm stuff show up in Sigmetrics. I think it could definitely get in, but somehow it gets sent other places (e.g. INFOCOM). Perhaps partially because the INFOCOM deadline is in between STOC/FOCS and Sigcomm/NSDI, whereas the Sigmetrics deadline basically corresponds to STOC & NSDI.

      I used to think the difference was in worst-case vs average-case analysis, but recently I’ve had papers in Sigmetrics on worst-case analysis… so I think it’s just a matter of people sending the papers! The deadline is in just over a week if you have something ready to go :)

  2. Pingback: Rigor + Relevance | What’s hot at EC and Sigmetrics this year?

  3. Pingback: Rigor + Relevance | Announcing NetEcon 2015

Leave a Comment

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s