ATC/OSDI 2025 impressions

This week I was in Boston for ATC/OSDI’25. Downtown Boston is a unique place where two/three-hundred-year-old homes and cobblestone streets are mixed with sleek buildings and biotech towers. The people here look wicked smart and ambitious (although lacking the optimism/cheer of Bay area people). It’s a sharp contrast from Buffalo, where the ambition is more about not standing out.

Boston was burning. 90°F and humid. I made the mistake of booking late, so I got the DoubleTree Boston-Downtown instead of the conference hotel. The mile-long walk to the Sheraton felt like a hike through a sauna. By the time I got there, my undershirt was soaked, and stuck to my back cold under the conference hall’s AC.  Jane Street's fitted t-shirt swag saved the day.

The Sheraton looked ragged from the outside, aged on the inside, but it was functional. The conference felt underfilled, with many empty seats. Later, I learned that the total ATC+OSDI attendance was under 500. That's a big drop from even ATC/OSDI 2022 attendance, which I discussed here.  

The conference was also low energy. Few questions after talks. People felt tired. Where were the 20+ strong MIT systems profs? ATC/OSDI happened in their backyard, but there was only one of them, and that only for the first day/morning.

The presentation quality was disappointing. A couple speakers looked like they were seeing the slides for the first time. Very demoralizing. Many ATC talks were just low-quality recorded videos. Visa issues accounted for many of the no-shows, which sucks. I can’t believe  we are dealing with this in 2025. Apparently, some European faculty are skipping U.S. conferences altogether now because of the political climate.

Missing speakers were somehow 10x more prevalent in ATC than OSDI. Two out of five talks were prerecorded (with no Q&A later) in the ATC session I chaired, as with several of the other sessions. I guess in a couple cases some US-based co-authors didn’t bother to show up and present. I just saw one OSDI talk being prerecorded. The session chair just told people to watch the recording later rather than playing the recording, which honestly felt like the right call. At the end of the day, I can still understand the prerecorded talks, but the low presentation quality in many of the talks are unexcusable. Boring talks meant empty seats, and the people in the room checking emails in their laptops rather than listening. 

The conference attendees this year had a striking bimodal distribution. On the one end, there were very young PhD students, and even some undergraduates. On the other: old-timers from the first USENIX days (70+ or 80+ years old), in town for USENIX's 50th anniversary and what feels like ATC’s last rites, as USENIX Annual Technical Conference was ended this year.

Conferences live and die by the community around them. When the community around the conference weakens, the quality and energy degrades. ATC is ended, and it seems like OSDI needs some TLC (tender loving care) to build up the community around it. This is a very hard thing to do, and I suspect there are no quick/easy hacks.

I don't know. Maybe people are tired and overwhelmed. The conference submissions keep going up by about 30% each year, the program committee reviewing is hard and thankless. It is getting harder and harder to get papers accepted. Maybe people are getting fed up with the paper publishing game, and as a result don't find conferences as useful or sincere. Oh, well...

Coincidentally, both NSDI and SOSP PC meetings start this Thursday, and there was a flurry of online discussion about the papers on Monday and Tuesday. Monday night I had to look back on papers I reviewed to respond to the rebuttals and discussion comments. I had to stay up late till after midnight, and I had Squid Games final season opened on hotel TV to get some background noise. Let's just say there are a lot of parallels with Squid Games and the publishing game.

I ran into many interesting folks in the hallway track: folks working on ML infrastructure, teaching LLMs to code, running AI infrastructure at OpenAI, researching new hardware for distributed systems. I met Amplify VC people, Sunil and Arjun, both very smart and technical. They fund distributed systems work and infrastructure for AI.

One pattern I noticed was there were lots of young folks skipping the PhD pipeline entirely. They went straight from school (with some undergraduate research work under their belt, and I presume good coding skills) to Anthropic or OpenAI. 

Hallway conversations should be easier to start. I always enjoy them once they get going. It’s the starting that's hard. But it’s worth pushing through the awkwardness and meet people.

Swag Rankings

I will talk about some intesting papers in a later blog post. Now, let’s get to the real reason we’re all here: the swag.

Databricks: A sticker. I swear that was it. They came with 20 databricks stickers to pass around. Are you serious?

Amazon: A shopping bag. Thanks, but no thanks.

Google: Stickers. And hats, but they only display the hats on the table, and don't give them to you. When I asked for a hat, they said it is for the students only (how did they know I am not a student?). And I don't know if they let students sign their soul out before passing them a hat.

Meta: Three ballpoint pens packaged/branded neatly. You give 100 millions a year to poach AI researchers, but you only pass around ballpoint pens at ATC/OSDI? (Well, on testing, the pens write real smooth, and my kids like them. Still beats the shopping bag, which I didn't bother picking up.)

Why send two staff to conferences just to hand out crap? You burned four days of salary and travel just to say, "Apply on our website" to people approaching the booth. This is not outreach. If you’re not going to hand out decent swag, don’t put up a booth. At least preserve some dignity.

Jane Street on the other hand was pure class. Their t-shirts are so soft and form-fitting, it feels like they were spun from distilled 401K pensions. You know what? I no longer feel bad about them recruiting top talent from research.