The Hyflow project is developing concurrency control abstractions, protocols, and mechanisms for current and emerging multicore architectures, cluster systems, and geographically distributed systems. A particular focus is to understand what concurrency control abstractions promote high programmability for these architectures, and how to support those programming abstractions with high performance, scalability, and dependability. A closely related focus is to build open-source experimental systems that embody the techniques, as well as incorporate them into existing open-source infrastructures.

The project has developed algorithms and mechanisms for the transactional memory concurrency control abstraction, transactional and concurrent data structures, safe memory reclamation, state machine replication techniques, and scalable consensus algorithms for geographically distributed systems. A recent focus is on the Byzantine failure model - an increasingly important failure scenario for cloud infrastructures - and developing consensus algorithms with high performance and scalability for that model.



Elpis - Generalized Consensus for Practical Fault Tolerance

Latency of Elpis, a multi-leader Cross Fault Tolerant consensus algorithm, and its competitors including XFT, a single-leader Cross Fault-Tolerant consensus algorithm (OSDI'16) and M2Paxos, a multi-leader Crash Fault-Tolerant consensus algorithm (DSN'16), on a key-value benchmark. For details, see our MIDDLEWARE 2019 paper.

Hyaline Hyaline
Performance of different variants of Hyaline compared to its competitors including Epoch-Based Reclamation and IBR (PPoPP'18) on the Bonsai tree data structure and Hyaline-S's ability to bound the number of unreclaimed objects. For details, see our PODC 2019 paper.



This work is supported in part by US National Science Foundation under grants CNS 0915895, CNS 1116190, CNS 1130180, and CNS 1217385, and AFOSR under grants FA9550-14-1-0163 and FA9550-14-1-0143. Any opinions, findings, and conclusions or recommendations expressed in this site are those of the author(s) and do not necessarily reflect the views of the National Science Foundation or AFOSR.