Dmitrii Ustiugov (@dmitriiustiugov) 's Twitter Profile
Dmitrii Ustiugov

@dmitriiustiugov

Assistant Prof at NTU Singapore. Cloud and serverless research with vHive. All opinions are mine.

Follow us at @studentssystems

ID: 2647238390

linkhttps://ustiugov.github.io calendar_today15-07-2014 09:06:31

391 Tweet

728 Followers

259 Following

Ion Stoica (@istoica05) 's Twitter Profile Photo

The progress in AI is down to three basic resources; (1) people (experts), (2) data, and (3) infrastructure. Arguably, at this point the US is only ahead in (3). Also, at this point the Chinese open source models are ahead. Not only DeepSeek but also Qwen. This is a fact.

Dmitrii Ustiugov (@dmitriiustiugov) 's Twitter Profile Photo

Looking for research topics in AI systems? Our latest SIGOPS article explores trends in scalable and autoscaling AI systems, providing numerous pointers to SOTAs and essential research tools. Let us know what you think in the comments below! NTU College of Computing and Data Science (CCDS) School of Informatics, The University of Edinburgh

DeepSeek (@deepseek_ai) 's Twitter Profile Photo

🚀 Day 6 of #OpenSourceWeek: One More Thing – DeepSeek-V3/R1 Inference System Overview Optimized throughput and latency via: 🔧 Cross-node EP-powered batch scaling 🔄 Computation-communication overlap ⚖️ Load balancing Statistics of DeepSeek's Online Service: ⚡ 73.7k/14.8k

Marios Kogias (@marioskogias) 's Twitter Profile Photo

Today at ACM PPoPP (conf.researchr.org/home/ppopp-2025) my student Scofield Z.Q. Liu will present our work on DORADD, an RPC runtime for deterministic parallel execution. Joint work with Matthew Parkinson and Musa Unal. Paper: marioskogias.github.io/docs/doradd.pdf Code: github.com/doradd-rt

Ana Klimovic (@anaklimovic) 's Twitter Profile Photo

LLM engineers often manage training data with ad hoc scripts and filesystem directory structures, which is tedious and prone to errors. We are releasing Mixtera, a lightweight, declarative data plane for dynamically selecting and mixing data of different modalities on the fly.

Dmitrii Ustiugov (@dmitriiustiugov) 's Twitter Profile Photo

We are releasing a new version of vHive v1.8: 📷 Ubuntu 24.04 support. Note the change in the recommended profile on CloudLab in the quickstart guide. 📷 Fixed the issue with resizing the root filesystem on CloudLab after reimaging the node. 📷 Updates in the networking setup.

Andrej Karpathy (@karpathy) 's Twitter Profile Photo

I wrote a quick new post on "Digital Hygiene". Basically there are some no-brainer decisions you can make in your life to dramatically improve the privacy and security of your computing and this post goes over some of them. Blog post link in the reply, but copy pasting below

I wrote a quick new post on "Digital Hygiene".

Basically there are some no-brainer decisions you can make in your life to dramatically improve the privacy and security of your computing and this post goes over some of them. Blog post link in the reply, but copy pasting below
Dmitrii Ustiugov (@dmitriiustiugov) 's Twitter Profile Photo

Join us at SESAME'25 at ASPLOS EuroSys 2023 on March 31st on the serverless cloud & AI systems! Fantastic schedule opened w/ a keynote by Peter Pietzuch Peter Pietzuch and followed by a panel and a series of workshop & tutorial talks on the hot topics. sesame25.github.io

Dmitrii Ustiugov (@dmitriiustiugov) 's Twitter Profile Photo

Enjoying ASPLOS & EuroSys 2025 ? If you are seeking a postdoc position with my lab focusing on serverless clouds and serverless AI, please DM me or let's have a chat! P.S. NTU Singapore also has many open faculty positions!

Dmitrii Ustiugov (@dmitriiustiugov) 's Twitter Profile Photo

Can’t improve what can’t measure! Submit papers with your brightest insights into computing system workloads. Deadline: June 21, 2025 Webpage: iiswc.org/iiswc2025/cfp.…

Can’t improve what can’t measure! Submit papers with your brightest insights into computing system workloads.

Deadline: June 21, 2025
Webpage: iiswc.org/iiswc2025/cfp.…
Baris Kasikci (@bariskasikci) 's Twitter Profile Photo

Efficiently running large Mixture-of-Experts (MoE) LLMs with constrained GPU memory is a critical challenge. We present Fiddler, a novel inference system that enables low-latency MoE inference through strategic CPU-GPU orchestration. Come see our work at #ICLR2025 ICLR 2026

Efficiently running large Mixture-of-Experts (MoE) LLMs with constrained GPU memory is a critical challenge. We present Fiddler, a novel inference system that enables low-latency MoE inference through strategic CPU-GPU orchestration. Come see our work at #ICLR2025 <a href="/iclr_conf/">ICLR 2026</a>
Tianyin Xu (@tianyin_xu) 's Twitter Profile Photo

A petition to SIGOPS to adopt the USENIX Annual Technical Conference (ATC) and retain its steering committee docs.google.com/document/d/1wK… (not sure whether it can be done by SIGOPS alone, but it's great to let the voice be heard)