News & Events

A very strong publication regord by the lab this year!

1. Congrats to Alon and Igor for having the NeurlLPM paper on hardware accelerator for Longest Prefix Match using Neural Nets accepted to MICRO'56 -- our very first lab publication in this super prestigious hardware venue!
2. Congrats to Sajy and Alon for the acceptance of P4-based virtual machine for Programmable Switches to NSDI'24, on the FIRST attempt, and the very first Sajy's paper! Also, this is Alon's 3rd NSDI paper, wow!
3. Congrats to Shai Bergman and our collabs from Imperial for getting the very cool Translation Pass Through paper on a new approach to Virtual Memory translation in Virtual Machines
4. Together with our collaborators from Microsoft Research Cambridge, KU Leuven and Intel we got 2 papers published in  USENIX Security and IEEE S&P. The work with Intel made it to the SGX ISA, the one with Microsoft is actively being used in production!

This year Prof. Silberstein will be teaching a course on building accelerated system in HiPEAC summer school (ACACES) More details: https://www.hipeac.net/acaces/2022

On May the 15th we were very happy to host a lab exposure event for undergraduate and prominent students considering graduate studies in the lab. Prof. Silberstein shared his research vision and our students and alumni shared their experiences doing research at the lab.
more details

Two papers by Shai and our collaborators got accepted! One talks about the OS Virtual Memory subsystem optimized for disaggregated memory. The second shows how OS swap can be significantly accelerated with the help of hardware-software co-design that leverages new ZNS SSD interfaces.

Congratulations to Alon Rashelbach, Lior Zeno, Haggai Eran, Igor de Paula and our collaborators for the acceptance of three papers authored by ACSL to NSDI'22!  Alon showed how to accelerate OpenV Switch using RQRMI neural nets. Lior, Alon and Igor showed how to build a distributed shared memory abstraction in programmable switches. Haggai showed how to virtualize congestion control in data centers.

Two papers accepted to ASPLOS'22: Revisor -- a system for automatic detection of speculative execution vulnerabilities on black-box CPUs: jointly with Microsoft Research Cambridge and TU Dresden, led by Oleksii Oleksenko and FlexConnect -- an infrastructure for direct p2p management of NICs from FPGAs: jointly with NVIDIA Networking (former Mellanox), led by Haggai Eran Congrats to the authors!!

Menachem's paper on systematically accelerating DNN training by using approximate tensor operations got into NeurIPS'21. Congratulations!

Michio is going to e-visit ACSL on the 11th of Feb. Here's his talk: Designing storage stacks for distributed, heterogeneous systems Storage software stacks are ubiquitous in computing systems. They are either embedded in individual hosts or disaggregated into a pool of servers or even dedicated racks. Although the performance of many storage systems has been largely constrained by slow disk devices, modern storage devices, such as NVMe SSDs and persistent memory, exhibit software overhead that impacts end-to-end performance. This talk introduces several instances in this problem space, and discusses solutions based on emerging opportunities.  

Lior Zeno got his first paper on Distributed Shared Memory abstraction for programmable switches (SwishMem) accepted to HotNets'20!

The paper "A computational approach to packet classification" by Alon Rashelbach, Ori Rottenstreich and myself has been accepted to SIGCOMM 2020! The paper introduces a novel Range-Query Recursive Model Index (RMI) which uses hierarchical neural network inference for packet classification and gains significant speedups over state of the art. Congrats to Alon!!