MACS Project Meeting, September 2016

Date:
Friday, September 9, 2016

Location:
MIT Stata Center, Star conference room (32-D463)

Schedule:

Talk abstracts:

High-Throughput BFT Protocols, Haibin Zhang

I will describe the design and implementation of BChain, a Byzantine fault-tolerant (BFT) state machine replication protocol, which performs comparably to other modern protocols in fault-free cases, but in the face of failures can also quickly recover its steady state performance. Building on chain replication, BChain achieves high throughput and low latency under high client load. At the core of BChain is an efficient Byzantine failure detection mechanism called re-chaining, where faulty replicas are placed out of harm’s way at the end of the chain, until they can be replaced. Our experimental evaluation confirms our performance expectations for both fault-free and failure scenarios.

Privacy-Preserving Smart Contracts, Ranjit Kumaresan

Cryptographic technologies such as encryption and authentication provide stability to modern electronic commerce. Recent technologies such as Bitcoin have the potential to further enhance the way we conduct electronic commerce. In this talk, I will describe my work in developing a robust theory of privacy-preserving contracts that are self-enforcing and do not require third-party intervention. Starting from Bitcoin-inspired abstractions, the theory enables building provably secure and scalable applications on top of cryptocurrencies.

Integrating MPC in Big Data Workflows, Malte Schwarzkopf and Nikolaj Volgushev

Secure multi-party computation (MPC) allows multiple parties to jointly compute on private input data. Existing MPC frameworks (e.g., VIFF, ShareMind) reduce the barrier-to-entry for MPC, but (1) still require significant domain knowledge; (2) integrate poorly with existing data analytics stacks and workflows; and (3) scale poorly to large data sets as they do not support efficient parallel processing.

In this work, we are developing a new, MPC-aware big data workflow manager based on Musketeer [2] to address these challenges. Our system assumes no familiarity with MPC and only requires the computation to be specified in straight SQL. It automatically determines the subsets of the workflow in which data cross trust domain boundaries, computing these subsets under MPC. It also automatically generates all MPC code, and pipelines it with steps of the workflow that execute in local, data-parallel systems such as Hadoop and Spark. The key challenge is to use “just enough” MPC to remain secure, but to avoid the performance penalty of running operations under MPC that do not require it because the operate purely on private data.

We have implemented an initial proof-of-concept with encouraging results [1], and are currently exploring several future research avenues that we briefly discuss in this talk.

[1] — Volgushev, N., Schwarzkopf, M., Lapets, A., Varia, M., and Bestavros, A. “Demo: Integrating MPC in Big Data Workflows”. In Proceedings of CCS 2016 (to appear).
[2] — Gog, I., Schwarzkopf, M., Crooks, N., Grosvenor, M.P., Clement, A., and Hand, S. “Musketeer: all for one, one for all in data processing systems” In Proceedings of EuroSys 2015.