with apologies

Sabbatical Diary: Weeks 35—38

· 5 min read · June 26, 2026 · #academic #cambridge #sabbatical #research #travel

Well. Apologies for another hiatus—that was a very busy few weeks.

I’ll put to one side some personal matters I’ve had to deal with, and some external work that was quite interesting.

Ph.D. supervision continued of course, with everyone showing some good signs of progress. Though disappointingly both submissions to SOSP got some pretty poor reviews. I might write more about that at some point—I’m clearly increasingly badly calibrated as to what counts as “novel” or “useful” when it comes to systems research results.1 Also interviewed some Ph.D. candidates.

Starting and then finished finalizing an M.Phil. project dissertation that I’ve been supervising was more fun. Extending Amjad’s work on fixing the CFS scheduler when scheduling cgroups, Jan Pytel did two things in response to reviews we’ve previously received. First, he demonstrated that the same problems occur on ARM and small devices by recreating them on a Raspberry Pi—relevant for those looking to run serverless workloads at the edge. Second, recreating a hybrid approximation of Amjad’s CFS-LAGS and the scx_flatcg scheduler in the kernels’ relatively recent eBPF sched_ext framework. It wasn’t possible to implement exactly the same approach as CFS-LAGS as the right hooks aren’t there, but he showed that comparable improvements could be obtained using hooks that were there. I thought it was nicely done anyway!

Spent some brief time planning the next UK Systems workshop. I really enjoy this one—it’s a fun crowd, it’s very informal, and it’s very supportive (or so I think anyway!) of early career researchers and Ph.D. students. Long may it continue to thrive.

Lyon train station, at night.
Lyon train station, at night.
The presentation room during a break at WOWMOM'26.
The presentation room during a break at WOWMOM'26.

And then the main event, and one of the two key reasons I’ve been swamped: I finally got round to writing a draft of my WOWMOM 2026 keynote before2 giving it as an SRG seminar.3 I think it went ok—the room was relatively full4 and the audience were quite engaged: got a good 30 minutes of Q&A and some very helpful feedback5 :)

via Maggiore, Bologna, at night.
via Maggiore, Bologna, at night.
A colonnade in Bologna.
A colonnade in Bologna.
The Two Towers of Bologna, not Mordor.
The Two Towers of Bologna, not Mordor.

And then, finally finally, I travelled to Bologna, Italy by train6 for WOWMOM. The talk got updated in light of the excellent feedback, so was ready at least several minutes before I had to give it. Unfortunately, like the incompetent muppet that I am, I setup to record it all, had everything ready to go, and then completely forgot to actually hit record. Sigh. Here are the slides anyway. Had some great followup conversations with Paolo Bellavista and some of his team—it seems there may be good overlap with the AI-RAN and FL work my group is doing here, so I’ll be following up on that.

A hotel staircase in Turin from below.
A hotel staircase in Turin from below.

And then I got back here to find that the weather was just as unpleasantly hot, and I had the inevitable catching up to do. Plus a very constructive conversation about the building energy consumption work I’m trying to get going – watch this space :)

  1. I mean, I thought that demonstrating how scheduler tweaks could allow the same workload to be carried out by a cluster with 20% fewer machines would count as both novel and useful. Or that, having apparently convincingly demonstrated the significant negative impact of an unaddressed problem in sampling for federated learning, solving that problem might be thought both useful and novel. But hey, who am I to say. ↩

  2. I say before. What I mean is, I was editing the slides until enough people arrived in the room that I really had to stop so that I could start speaking from the beginning. After 25 years you’d think I’d have learned. But clearly not. ↩

  3. https://talks.cam.ac.uk/talk/index/256490/ ↩

  4. Full disclosure: it was a much smaller room than we usually use because the exams were taking place in the usual room. ↩

  5. Particularly Stephen Kell’s written feedback which I found very useful. Props particularly to Michael Schapira, Jon Crowcroft and Ian Leslie for also giving some very useful comments. ↩

  6. Out via Cambridge—St Pancras—Lille—Lyon–|–Milan—Bologna and back via Bologna—Turin–|–Paris—St Pancras—Cambridge ↩