HyFlow


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 (e.g., Infinispan).

Ongoing efforts are exploring high performance fault-tolerant transactional memory for cluster systems, high performance software transactional memory for multicore architectures, scalable transactional data structures, and scalable transactional protocols for geographically distributed systems.

 


Research Areas

  • Scalable Transactional Memory

  • Transactional Data Structures

  • Fault-Tolerant Transactions for Multicore and Distributed Systems

  • Geo-Replicated Transactions

HiperTM

Performance of HiperTM deployed on 8 nodes running Bank benchmark. HiperTM is a fault-tolerant software transactional memory system for clusters, which uses active replication, yielding full failure masking. For details, see our ICDCN 2014 paper.

 
Optimistic Transaction Boosting

Performance of Optimistic Boosted Skip-List, with 64K objects and 5 operations per transaction, compared to the original boosted Skip-List and the Lazy (non-transactional) Skip-List. Boosting enables transactional versions of data structures, which yields composability. For details, see our PPoPP 2014 (poster) paper and our project web site.

Selected Publications

All papers

 
This work is supported in part by US National Science Foundation under grants CNS 0915895, CNS 1116190, CNS 1130180, and CNS 1217385. 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.