Asaf Cidon

1/ Graph-based vector search spends roughly twice as many I/Os retrieving the 20th-best result as the first, even though the first is usually the one that matters. Our MLSys 2026 paper with @jiananlu_nt (@Princeton) and @asafcidon (@Columbia) argues this is a metric problem before it is a systems problem. A thread, paper link below 👇

@lauriewired @zhong_yuhong @talzuss @yanniszark XRP could also serve the purpose of bandwidth reduction (eg pushing down filters). Our original use case was bypassing the kernel storage stack, where bandwidth is less of a concern, but rather triggering many I/Os. But for programmable storage bandwidth reduction make sense.

nice! just read through the XRP paper. looks like it scaled really well! in the conclusion it's mentioned envisioning it as a potential common interface. do you think commodity drives would ever have enough compute on device to be worth it? To be specific, in my post I was thinking more of a bandwidth reduction play on-device, whereas XRP is more of a latency play on-host.

if you’re a CS/EE student write your thesis on JIT compilation of eBPF for NVMe controllers there’s huge career alpha in computational storage; the standards are *just* starting to exist (TP4091)

@DimitriosMitsos @talzuss @linuxplumbers Sounds cool! Would be nice to implement chameleon it on cache_ext :-)

@asafcidon @talzuss @linuxplumbers Hey there Asaf, i think you might be intrested in this x.com/DimitriosMitso…

My PhD student @talzuss presented our work on cache_ext at @linuxplumbers, a new framework that allows applications to customize the OS page cache via eBPF. Imagine giving a database full control of page cache eviction/admission/prefetching policies. Hoping to get it upstreamed!

Last week I presented cache_ext at SOSP'25! 🎉 Linux's page cache uses an inflexible LRU-like eviction policy for all applications, leaving performance on the table. cache_ext allows users to customize the page cache using eBPF-based policies to matter match their workloads. 🐝

Super excited to announce that cache_ext is accepted in SOSP 2025! 🎉 The Linux page cache is a central component of the kernel, yet it's been using the same policy (2Q-LRU) for decades, with only some recent improvements (e.g., MGLRU from Google).

Every single one of the 11 Meta superintelligence hires is an immigrant who did their undergrad abroad. 7 China, 1 India, 1 Australia, 1 UK, 1 South Africa. 8 are PhD or PhD dropouts in the US. Immigration is key to US AI innovation.

@MathiasPoertner @LSEGovernment @LSEnews Congrats Mathias!!

Exciting news to share — I have been awarded tenure and promoted to Associate Professor @LSEGovernment @LSEnews! I’m immensely grateful for the support and guidance from colleagues, mentors, and students, and I’m looking forward to the journey ahead!

@MarkLevineNYC Run for mayor instead!

Big news: Later today I'll be officially announcing my candidacy to be the next Comptroller of NYC. I will use the powers of the office to address our city's historic crisis of affordability and livability, and to restore broken confidence in government. Join us!

@pramod_bhatotia Agreed!! We should be accepting all those “borderline” papers for whom the final acceptance/rejection decision is completely random, and stop wasting all of our collective time (students, faculty, reviewers) in the endless resubmission cycle

@jaewon_chung_cs Congrats @jaewon_chung_cs! Really cool work

Perseus was accepted to appear at SOSP'24! Perseus is an energy optimization system that identifies and reduces *energy bloat* form large model training. I wrote up a quick blog post: ml.energy/zeus/research_… 1/6

“POTUS thought he was past his best and, for the greater good, should resign and hand off to someone younger and more able” Tenured profs:

@ColumbiaCompSci @CUSEAS @EE_ColumbiaSEAS

Paper link: usenix.org/system/files/o…

Proud advisor moment #osdi24. @zhong_yuhong presented our work on managing CXL in virtualized cloud environments