RIP Usenix ATC(bcantrill.dtrace.org)
bcantrill.dtrace.org
RIP Usenix ATC
https://bcantrill.dtrace.org/2025/05/11/rip-usenix-atc/
39 comments
> Computers and the Internet started with academics.
Only if one categorizes academics as a research arm of military funded research goals. Not that there is anything wrong with that, but let's be intellectually honest.
> UNIX is a product of a quasi-academic research lab.
A wholly owned subsidiary of the commercial entity AT&T as far as the UNIX trademark is concerned. The BSD family would have something to say about Unix (note case difference) however.
> The Center of Gravity shifted in the late 90s/early 2000s, I reckon. Real world needed distributed systems ...
Distributed systems and the networking protocols now assumed (TCP and UDP) was a battle fought before the late 90's. See Novell[0] for an alternative many espoused at the time and Sun's "The Network is the Computer" mantra from the mid 80's.
> Not saying USENIX didn’t have factions. I just question the word “capture” about academics, who invented all this stuff before there was money in it.
There was always "money in it." Academic funding is, by definition, justifying the allocation of money for the advancement of knowledge often pitched as relevant to the benefactor. Again, not that there is anything wrong with this.
0 - https://en.wikipedia.org/wiki/Novell
Only if one categorizes academics as a research arm of military funded research goals. Not that there is anything wrong with that, but let's be intellectually honest.
> UNIX is a product of a quasi-academic research lab.
A wholly owned subsidiary of the commercial entity AT&T as far as the UNIX trademark is concerned. The BSD family would have something to say about Unix (note case difference) however.
> The Center of Gravity shifted in the late 90s/early 2000s, I reckon. Real world needed distributed systems ...
Distributed systems and the networking protocols now assumed (TCP and UDP) was a battle fought before the late 90's. See Novell[0] for an alternative many espoused at the time and Sun's "The Network is the Computer" mantra from the mid 80's.
> Not saying USENIX didn’t have factions. I just question the word “capture” about academics, who invented all this stuff before there was money in it.
There was always "money in it." Academic funding is, by definition, justifying the allocation of money for the advancement of knowledge often pitched as relevant to the benefactor. Again, not that there is anything wrong with this.
0 - https://en.wikipedia.org/wiki/Novell
I think what they're saying is this:
USENIX ATC at its best was always focused on operations and real use of Unix. It was the Unix User's Group.
But academic CS has a shortage of real publishing opportunities, causing less immediately practical things to flood USENIX ATC and smother it.
USENIX ATC at its best was always focused on operations and real use of Unix. It was the Unix User's Group.
But academic CS has a shortage of real publishing opportunities, causing less immediately practical things to flood USENIX ATC and smother it.
Academic CS has no shortage of prestigious or even good publishing venues. Just within systems and networking, there is SOSP, OSDI, NSDI, EuroSys, PODC, DISC, FAST, SIGCOMM, CIDR. If you extend to databases there's also VLDB, SIGMOD, etc. These are all Tier-1/Tier-1.5 venues.
Academic CS didn't take over ATC because of a shortage of venues.
Academic CS didn't take over ATC because of a shortage of venues.
You made this comment twice, so I guess I'll reply to it twice, but the reality is that the acceptance rates of ATC was very, very low (again, under 13% in 2004). Low acceptance rates actually do indicate a shortage of good publishing venues (certainly relative to the number of submissions), but it would be interesting to look at the ATC acceptance rate over time; if it was much (much) higher in its final years, it would be easier to accept your assertion.
17% in 2024. More submissions than other conferences. Attendance numbers probably skewed by cross-registration with OSDI.
https://github.com/emeryberger/csconferences indicates that ATC acceptance rate has averaged 20% in the last five years, so better than ~13%, but not by too much.
However, at least speaking as an academic, I wouldn't say that ~20% acceptance rate is necessarily indicative of a shortage of good venues. There is plenty of not-so-good research that is submitted to top places that has no hope of getting in. (My experience is from computer security research, where the acceptance rate of security conferences has gone down, but the fraction of good papers has also actually gone down, so the fraction of good papers getting in has roughly stayed the same.)
That being said, ATC seems to indeed have been a high-prestige conference back in the day, and hence indeed competitive. My experience is from recent years, where it was viewed as a good-but-not-tier-1 conference in systems research.
However, at least speaking as an academic, I wouldn't say that ~20% acceptance rate is necessarily indicative of a shortage of good venues. There is plenty of not-so-good research that is submitted to top places that has no hope of getting in. (My experience is from computer security research, where the acceptance rate of security conferences has gone down, but the fraction of good papers has also actually gone down, so the fraction of good papers getting in has roughly stayed the same.)
That being said, ATC seems to indeed have been a high-prestige conference back in the day, and hence indeed competitive. My experience is from recent years, where it was viewed as a good-but-not-tier-1 conference in systems research.
IMO the implication that the "in-person conference" nature of the forum is what led to its demise rather than its capture by academics/abandonment by industry practitioners is incorrect. I think a better summary of the cause is that most academic research is not that relevant to industry but academics need the prestige of forums like Usenix ATC in a way that industry does not which naturally leads to squeezing industry (and thus much of the relevance) out of the forum over time. I think the same would have happened to a virtual conference/printed journal/etc.
I definitely agree with you, and apologies if that didn't come through in the piece! (Perhaps the rare case where I undershot on the metaphors?)
One thing I didn't mention but certainly believe: academics were attracted to USENIX ATC because of its attendance count (especially at the height of the Dot Com boom when it had nearly 2,000 attendees!) -- but no one really took apart who was attending or what they were looking for. So the conference became more academic because of the attendee count -- but that it became more academic also drove the attendee count down. (I heard from a lot of practitioners who attended that they struggled to find sessions that were relevant to their work in even the broadest sense.) I know I link to it in the piece, but I think Rik Farrow's piece[0] really got right to the heart of all of this.
[0] https://www.usenix.org/system/files/login/articles/login_fal...
One thing I didn't mention but certainly believe: academics were attracted to USENIX ATC because of its attendance count (especially at the height of the Dot Com boom when it had nearly 2,000 attendees!) -- but no one really took apart who was attending or what they were looking for. So the conference became more academic because of the attendee count -- but that it became more academic also drove the attendee count down. (I heard from a lot of practitioners who attended that they struggled to find sessions that were relevant to their work in even the broadest sense.) I know I link to it in the piece, but I think Rik Farrow's piece[0] really got right to the heart of all of this.
[0] https://www.usenix.org/system/files/login/articles/login_fal...
I didn't mean to imply that I disagreed with anything other than that specific point (which perhaps is more nostalgia for the halcyon days of in-person conferences than anything else).
I do think that you correctly identified a major source of the problem back in 2004 as economic factors. I was at Mozilla when Rust was built (though not directly involved) and the sum of Mozilla's investment into Rust over the years easily broke into the 8 digits. Google's investment in Go I'm sure is an order of magnitude or two more than that. This is simply beyond the capacity of any academic institution or grant process. The only academic efforts that get to this level require Acts of Congress (e.g. LIGO, the Human Genome Project, the James Webb Space Telescope, etc).
And anyone who can afford to drop $10M+ into development can find channels for distributing and publicizing their work that don't go through program committees. Open source is definitely a big part of that but I don't think that's the whole story. I'd certainly count CUDA as a major advance in systems software since 2004, for instance.
I do think that you correctly identified a major source of the problem back in 2004 as economic factors. I was at Mozilla when Rust was built (though not directly involved) and the sum of Mozilla's investment into Rust over the years easily broke into the 8 digits. Google's investment in Go I'm sure is an order of magnitude or two more than that. This is simply beyond the capacity of any academic institution or grant process. The only academic efforts that get to this level require Acts of Congress (e.g. LIGO, the Human Genome Project, the James Webb Space Telescope, etc).
And anyone who can afford to drop $10M+ into development can find channels for distributing and publicizing their work that don't go through program committees. Open source is definitely a big part of that but I don't think that's the whole story. I'd certainly count CUDA as a major advance in systems software since 2004, for instance.
I don't feel like I have a good understanding of all the context but I don't think that's the implication at all. The post does offhandedly mention that doing online only conferences get "more bang for the buck" but the rest of the post and linked articles and talks discuss other aspects and, in my opinion, mirror your last sentiment about it happening to all conferences and journals.
For example, one of the linked talks from 2016 "A Wardrobe for the Emperor" [0] talks about adding a more social website features (stars, likes, comments, feedback, etc.) to arXiv to improve it. He also talks about the "Papers we love" meetups, which are in person.
I think the Usenix ATC is a highlighting a deeper problem and that it is happening across other conferences, journals and disciplines. I don't have a good theory of why. I think Cantrill's observations are part of the puzzle but I don't have a good sense for why things changed so drastically (and I think they have, compared to 20-40 years ago).
[0] https://www.youtube.com/watch?v=gAEiXWO44bQ&t=3457s
For example, one of the linked talks from 2016 "A Wardrobe for the Emperor" [0] talks about adding a more social website features (stars, likes, comments, feedback, etc.) to arXiv to improve it. He also talks about the "Papers we love" meetups, which are in person.
I think the Usenix ATC is a highlighting a deeper problem and that it is happening across other conferences, journals and disciplines. I don't have a good theory of why. I think Cantrill's observations are part of the puzzle but I don't have a good sense for why things changed so drastically (and I think they have, compared to 20-40 years ago).
[0] https://www.youtube.com/watch?v=gAEiXWO44bQ&t=3457s
I don't think academics view ATC as particularly highly prestigious. Just within systems there are many other conferences which are either more, or just as, prestigious as ATC: OSDI, SOSP, NSDI, SIGMOD, VLDB, SIGCOMM, EuroSys. Especially the first three are viewed to be much higher prestige than ATC, at least from my perspective.
I really don't care about how prestigious academics view (or viewed) ATC, but for whatever it's worth when our paper was accepted in 2004, was just under 13%. That is bluntly too oversubscribed for the practitioner (whose job is to ship systems, not win paper popularity contests -- or adjudicate those popularity contests by serving on program committees). In some ways, if ATC wasn't thought prestigious, it's even worse -- it means that practitioners lost our conference for nothing at all.
A final note: given that OSDI is on your list of prestigious conferences, the way the program committee was conducted in 2010 (which I outlined in my ATC keynote[0]) should be particularly galling.
[0] https://speakerdeck.com/bcantrill/a-wardrobe-for-the-emperor...
A final note: given that OSDI is on your list of prestigious conferences, the way the program committee was conducted in 2010 (which I outlined in my ATC keynote[0]) should be particularly galling.
[0] https://speakerdeck.com/bcantrill/a-wardrobe-for-the-emperor...
I used to be a grad student in a file systems lab at UC Santa Cruz over a decade ago. It was considered a dream to publish in OSDI or SOSP; our lab’s most successful paper was the OSDI 2006 paper on Ceph. We also thought very highly of VLDB, though most of our work was outside the interests of the database community. We regularly published in USENIX FAST and made occasional appearances at USENIX ATC.
I think ATC is viewed as being a prestigious conference at roughly the level of some of the above.
The big issue may be companies clamping down on divulging details regarding vulnerabilities and mitigations. Cybersecurity is significantly more weaponized now than it was in the 90s and companies might be facing immense pressure to limit the wrong kind of attention.
It was a fun conference. RIP.
Anybody remember the January 21–23 '87 Winter Usenix in Washington, DC, that got snowed in by the giant "Blizzard of Discontent" with 14 inches of snow that closed down Metro? The beloved/infamous Mayor Marion "The Bitch Set Me Up" Barry was attending the Super Bowl in California at the time, a fine tradition that Ted Cruz honored in Cancun during the February 2021 winter storm in Texas.
https://www.youtube.com/watch?v=HSRW18ahWG0
https://www.youtube.com/watch?v=r5KAw9MHeP0
https://www.youtube.com/watch?v=NZSROdXWTTc
https://www.youtube.com/watch?v=mTMif8cGlcE
Or the one after it on June 8–12 '87 Summer Usenix in Phoenix, Arizona, where they took everybody to a dude ranch for steak? My friend ||ugh Daniel was a vegetarian, and didn't want steak, so they told him to "git" into that line over "yonder", which he did. When he got to the front of the line, they plopped a half of a chicken onto his plate, like that's what a vegetarian wants to see and eat for dinner.
Anybody remember the January 21–23 '87 Winter Usenix in Washington, DC, that got snowed in by the giant "Blizzard of Discontent" with 14 inches of snow that closed down Metro? The beloved/infamous Mayor Marion "The Bitch Set Me Up" Barry was attending the Super Bowl in California at the time, a fine tradition that Ted Cruz honored in Cancun during the February 2021 winter storm in Texas.
https://www.youtube.com/watch?v=HSRW18ahWG0
https://www.youtube.com/watch?v=r5KAw9MHeP0
https://www.youtube.com/watch?v=NZSROdXWTTc
https://www.youtube.com/watch?v=mTMif8cGlcE
Or the one after it on June 8–12 '87 Summer Usenix in Phoenix, Arizona, where they took everybody to a dude ranch for steak? My friend ||ugh Daniel was a vegetarian, and didn't want steak, so they told him to "git" into that line over "yonder", which he did. When he got to the front of the line, they plopped a half of a chicken onto his plate, like that's what a vegetarian wants to see and eat for dinner.
I definitely remember that one, one of the first ATCs I attended. I took an NFS course and the speaker told us that after the class was over for the day, all of the Californians were planning to go out and play in the snow! I had planned to show off my brand new "DEV CAR" license plates at that ATC, but the snow was so bad I took Metro. I still have those tags.
Sounds like you needed "DEV PLOW" for DC!
And "DEV HORSE" for Phoenix.
And "DEV HORSE" for Phoenix.
BSDCan is happening in one month, and along with sister conferences, has always kept a good mix of industry and academic work imo. There's a few PhD or masters student projects, but many of the talks are practical industrial work, things that have shipped.
The flip side of this, though, is that it has probably never been easier to effectively distribute exciting new systems research, both because of the proliferation of smaller, more focused conferences with public video, and because we've all just gotten better at promoting stuff on the Internet. Usenix at its worst was a gatekeeper rather than a facilitator.
I absolutely agree, and I got to as much in my USENIX ATC 2016 keynote[0]: there are so many vectors now for sharing new ideas, it's almost hard to remember that in the heyday of USENIX ATC, technical conferences were really one of the only ways (along with USENET, really) for practitioners to broadcast a new idea. While there were perhaps upsides to having such limited vectors with respect to high signal, there were of course many more downsides; systems research has been well-served by information connectedness!
[0] https://speakerdeck.com/bcantrill/a-wardrobe-for-the-emperor...
[0] https://speakerdeck.com/bcantrill/a-wardrobe-for-the-emperor...
Related:
Usenix ATC Announcement
https://news.ycombinator.com/item?id=43933511
Usenix ATC Announcement
https://news.ycombinator.com/item?id=43933511
Thanks!
Usenix ATC Announcement - https://news.ycombinator.com/item?id=43933511 - May 2025 (10 comments)
Usenix ATC Announcement - https://news.ycombinator.com/item?id=43933511 - May 2025 (10 comments)
I think people are glossing over the fact that Usenix ATC wasn’t even a good academic conference. If you’re submitting high-quality academic work to Usenix, you’re sending it to one of the higher prestige Usenix conferences in a specific subfield, such as Usenix Security or NSDI.
I don't agree with this. ATC is a good systems conference.
This is not really a contradiction, IMHO. I published about systems 15 yrs ago and experienced the crisis in systems research. My personal take is, that the academic community failed to set new standards and research methods for computing system. (Maybe they cannot even exist in our short lived publish or perish world) I still review for conferences that still have systems (like ieee percom), but it is really difficult to reach consensus on system papers in comparison to AI or HCI stuff, that is also published.
I agree that for a lot of purposes online conferences are a great experience (p99 conf has been really fun for example) but nobody has really cracked the hallway track. I've heard about smaller conferences like Monktoberfest but their smallness being a main appeal necessarily limits their scope. Meetups seem to lack a certain je ne sais quoi. I wonder if someone will either figure out the online hallway track experience or if there is room for a new sort of in-person event?
I perceive a negative colouring of the "academic" notion in this post. And I totally get some of the pathologies implied: publish-or-perish is a harsh environment, lots of snobbery.
With all respect, however, this industry/academia dualism looks to me at best as a false and at worst as a harmful dichotomy. I mean, there are similarly cliché pathologies in whatever one chooses to overload "industry" with. There are successful academic projects that end up in industry (RISC-V) and successful industrial open-source (?) projects loved by academics (Rust). At the end of the day, everyone wants to build something. Tools and methods are needed. Some of them may not exist yet. The act of building one's tools and methods (or something completely new with existing tools and methods) is research regardless from where it happens or how it is labelled.
Why is this not enough? What extra benefits go with differentiating myself as either an academic or a practitioner? Honest question!
POV: CompEng PhD looking for industrial vacancies.
With all respect, however, this industry/academia dualism looks to me at best as a false and at worst as a harmful dichotomy. I mean, there are similarly cliché pathologies in whatever one chooses to overload "industry" with. There are successful academic projects that end up in industry (RISC-V) and successful industrial open-source (?) projects loved by academics (Rust). At the end of the day, everyone wants to build something. Tools and methods are needed. Some of them may not exist yet. The act of building one's tools and methods (or something completely new with existing tools and methods) is research regardless from where it happens or how it is labelled.
Why is this not enough? What extra benefits go with differentiating myself as either an academic or a practitioner? Honest question!
POV: CompEng PhD looking for industrial vacancies.
> successful industrial open-source (?) projects loved by academics (Rust)
Is Rust loved by academics? And much more importantly in my mind, was it even recognized by academics before it became an industrial success?
The very first published Rust paper that I'm aware of appeared in the "ACM SIGAda Ada Letters" (https://dl.acm.org/doi/10.1145/2692956.2663188). Today that paper is the most cited paper to ever appear in that journal (which has a 4 decade history) and it's not even close. It also has more citations that all but three papers that appeared at PLDI that same year, for comparison. That certainly doesn't suggest to me that it was recognized at the time. This was also published only 6 months before Rust shipped 1.0. It wasn't that early.
My (third-hand and now-decade-old, so take it for the very little it's worth) recollection is that academic forums weren't interested in Rust because nothing in it is particularly novel in a PL theory sense (a point Graydon himself made from the very beginning, see slides 6 and 7 at http://venge.net/graydon/talks/intro-talk-2.pdf). But it did package those ideas into something that was practically usable for industry and at this point the results speak for themselves.
IMO this is a good example of why a lot of "practitioners" in industry wouldn't bother trying to publish anything in a forum dominated by academics.
Is Rust loved by academics? And much more importantly in my mind, was it even recognized by academics before it became an industrial success?
The very first published Rust paper that I'm aware of appeared in the "ACM SIGAda Ada Letters" (https://dl.acm.org/doi/10.1145/2692956.2663188). Today that paper is the most cited paper to ever appear in that journal (which has a 4 decade history) and it's not even close. It also has more citations that all but three papers that appeared at PLDI that same year, for comparison. That certainly doesn't suggest to me that it was recognized at the time. This was also published only 6 months before Rust shipped 1.0. It wasn't that early.
My (third-hand and now-decade-old, so take it for the very little it's worth) recollection is that academic forums weren't interested in Rust because nothing in it is particularly novel in a PL theory sense (a point Graydon himself made from the very beginning, see slides 6 and 7 at http://venge.net/graydon/talks/intro-talk-2.pdf). But it did package those ideas into something that was practically usable for industry and at this point the results speak for themselves.
IMO this is a good example of why a lot of "practitioners" in industry wouldn't bother trying to publish anything in a forum dominated by academics.
Yes, all Rust papers anyone tried to submit were consistently rejected up until Rust became popular, at which point Rust became the hot new thing in applied programming language research. Academic PL is very insular (to its detriment, I'm convinced).
Damn. As someone who has dabbled in OS history research Usenix's archives were a God-send. I hope they continue to maintain them, and that other groups can take up the mantle of hosting cutting-edge computer science research (at least to the extent its still happening).
Building relationships and community is still essential, and it does require some in-person events. A central annual conference is time-efficient. How will these people get together?
Similarly, Stanford's EE380 has slipped from these mortal coils. Brian Cantrill spoke there as well. BITD, EE380 seemed like the Senate of Silicon Valley. Startups would announce at EE380. But then it went from relevance to thinly attended and taped for YouTube to archives.
This wasn't academic capture. There's a simpler explanation. No one goes to talks anymore.
This wasn't academic capture. There's a simpler explanation. No one goes to talks anymore.
Oh no! I had honestly never heard of EE380 when they asked me to present[0], but I really appreciated the fact that it was (either implicitly or explicitly?) open to the public. I wish I had heard of it earlier, and sorry to see it go!
[0] https://www.youtube.com/watch?v=vvZA9n3e5pc
[0] https://www.youtube.com/watch?v=vvZA9n3e5pc
It would be seriously disappointing if the world we are moving to is one where people don’t care about in person talks anymore. Not only do I love in-person conferences, I would probably stop caring about publishing papers if it was disconnected from the social parts. The social parts of PLDI, SPLASH, and yes, ATC (where I was fortunate enough to win a best paper on my one and only submission there) have been major highlights of my career, and more than once have led to friendships, interesting projects, and even employment. I was part of the team that helped make PLDI virtual during the pandemic in 2020, and as well as it went, I found it so painful to attend that I basically did not attend.
I understand that I am in the minority here when it comes to remote work, etc, but if opportunities to learn from and socialize in person with colleagues go away, then I will too. Woodworkers don’t seem so enamored with virtual this and that, so I would probably jump ship and do something else.
I understand that I am in the minority here when it comes to remote work, etc, but if opportunities to learn from and socialize in person with colleagues go away, then I will too. Woodworkers don’t seem so enamored with virtual this and that, so I would probably jump ship and do something else.
The Center of Gravity shifted in the late 90s/early 2000s, I reckon. Real world needed distributed systems, and neural networks, and crypto math, and hardware design, and …
There’s still a place for academic research but it’s a much different place in the Gilded CPU Age. It’s the god of the non-commercial gaps: for a while there if it seemed like it’d make money, someone .com giant with gigabucks of free cash had someone on it. Why work for a university if gigabucks.com is hiring?
Not saying USENIX didn’t have factions. I just question the word “capture” about academics, who invented all this stuff before there was money in it. It feels like retiring the Annual Technical Conference is the final step in Usenix’s capture by industry.