These are the old pages from the weblog as they were published at Cornell. Visit for up-to-date entries.

December 09, 2004

From Worlds to Vegas

I did a conference tour the past days. It started off with Usenix' Workshop on Real, Large Distributed Systems on Sunday. I saw a few reasonable papers but I was mainly surprised that for many in the research world real & large still appears to be planetlab. Which is a selection of 200 (or 400, depending on who you talk to) nodes placed at privileged places around the world. It is certainly not my definition of large and it is real insofar it exists, but it is far from my reality. The lone voice that somewhat resonated with my ideas was Greg Papadopoulos, Sun’s CTO, he was on the final panel together with Carl Kesselman from the Grid Forum and Larry Peterson from Princeton who leads the PlanetLab effort. Greg rightly observed that both the other panel members are pushing for more and more complex technologies, without any guarantees that these will result in systems that can function better at large scale in the real world, or even solve relevant problems in terms of real scale.

I had to miss most of OSDI, which followed WORLDS, because I had promised to give presentation at the Computer Measurement Group’s annual conference. This was the last of my Cornell commitments that carried over into time. The CMG audience is very targeted on tuning, capacity planning, etc. and quite a few are die-hard industrial professional, still involved with running mainframes. Very interesting folks and I enjoyed talking with them about their problems. I am not sure whether they actually appreciated my talk on web-services which seems somewhat remote from their core areas.

There was one thing that was a new experience for me. The attendees fill in little slips with speaker feedback. I thought they went back the organizers, but that is not the case. The “room monitor” collects the feedback forms and gets them to the speaker, so I ended up with 40-50 little notes with feedback about the talk or my presentation skills. Very useful.

Jonathan Hardwick has a number of excellent posts from the CMG conference

In the evening I strolled around Las Vegas a bit, rode the bus to the north end of the strip and back to the south end. After I recovered from image overload that all the big casinos shoot at you, I am surprised how obvious the human misery is that seems to be the underlying current in Las Vegas. It is so obvious that it is all an illusion that it makes you wonder why we actually expose ourselves to it. And just outside the main casino area the presence of poverty and high-crime are so pervasive that the contrast is huge. But even though I was happy to leave town today, I do want to come back to take a bit longer look at it.

Posted by Werner Vogels at December 9, 2004 01:34 PM


Hi Werner. I agree in principle with your comment on scale (and in fact tried to make that point in my talk). But how can we academics get access to *actually* real, large distributed systems, short of taking a "sabbatical" of some sort at a company like Amazon, eBay, Google, etc.?

I did a paper several years ago studying failures in large Internet services. It was like pulling teeth to get folks to allow me access to their data. After six months of work, I got exactly 3 out of the 40+ companies I approached, to agree.

Do you have any suggestions on how academics can more effectively collaborate with the folks who architect, implement, and operate *actually* large, distributed systems?

Posted by: David Oppenheimer on December 9, 2004 06:11 PM