I have returned to Switzerland after slightly more than 2 years (02/2012-05/2014) as a postdoctoral associate in the Department of Computer Science at Cornell University in Ithaca, New York. During this time, my research focused on distributed systems, particularly on scalability and energy efficiency aspects of fault-tolerance. I have also collaborated with the Nuprl group on using formal methods to build robust replication protocols.
nschiper at cs dot cornell dot edu
455 Gates Hall
Ithaca, NY 14853
My PhD work focused on developing scalable protocols to replicate data over wide area networks. This work resulted in developing several atomic multicast protocols that ensure agreement on the sequence of messages delivered and offer good scalability properties. On top of this multicast service we built P-Store, a replicated key-value store that offers full transactional support and any data sharding scheme. More recently, I started investigating the performance of various replication protocols when data cannot be sharded easily.
Collaborators: Fernando Pedone and Pierre Sutra.
According to a report by J. G. Koomey, data centers globally consumed more than 1% of the worldwide electricity use in 2010. This project looks at ways to improve the energy efficiency of replication protocols, a very important building block to mask common failures in modern clouds. In the context of a replicated database, energy efficiency is measured as the number of transactions committed per Joule, or equivalently, the throughput of committed transactions per Watt. We have measured the efficiency of common replication approaches and propose new ways to improve this efficiency in a DSN 2014 paper.
Collaborators: Fernando Pedone and Robbert van Renesse.
Correct-by-construction Replication Protocols
In an effort to develop more robust fault-tolerant software we have built ShadowDB, a replicated database whose failure-handling code has been proven correct. To increase failure-independence of the replicas, ShadowDB offers automatic and correctness-preserving means to diversify their implementations. Preliminary results appeared at HotDep'12 and WRiPE'12. A more complete presentation of this project has been accepted at DSN 2014 and is available for download here. ShadowDB is part of the CRASH initiative.
Collaborators: Vincent Rahli, Robbert van Renesse, Mark Bickford, and Robert L. Constable.
Enhancing the Dependability of Distributed Applications
A substantial proportion of application bugs lead to non-crash failures: data corruption, silent errors, and unauthorized user access are some of the consequences of these bugs. This project aims at proposing BFT replication protocols that are simpler to implement, reduce hardware costs, and offer good performance. Examples of such protocols are Byzantine Chain Replication and BFT-DUR.
Collaborators: Robbert van Renesse and Fernando Pedone.
EventsPC member of the 15th International Symposium on Stabilization, Safety, and Security of Distributed Systems SSS 2013
Shadow PC member of the 2013 European Conference on Computer Systems Eurosys 2013
PC member of the 16th International Conference On Principles Of Distributed Systems OPODIS 2012
PC member of the 12th International IFIP Conference on Distributed Applications and Interoperable Systems DAIS 2012