Making BFT Protocols Really Adaptive

Many state-machine Byzantine Fault Tolerant (BFT) protocols have been introduced so far. Each protocol addressed a different subset of conditions and use-cases. However, if the underlying conditions of a service span different subsets, choosing a single protocol will likely not be a best fit. This yields robustness and performance issues which may be even worse in services that exhibit fluctuating conditions and workloads. In this paper, we reconcile existing state-machine BFT protocols in a single adaptive BFT system, called ADAPT, aiming at covering a larger set of conditions and use-cases, probably the union of individual subsets of these protocols. At anytime, a launched protocol in ADAPT can be aborted and replaced by another protocol according to a potential change (an event) in the underlying system conditions. The launched protocol is chosen according to an 'evaluation process' that takes into consideration both: protocol characteristics and its performance. This is achieved by applying some mathematical formulas that match the profiles of protocols to given user (e.g., service owner) preferences. ADAPT can assess the profiles of protocols (e.g., throughput) at run-time using Machine Learning prediction mechanisms to get accurate evaluations. We compare ADAPT with well known BFT protocols showing that it outperforms others as system conditions change and under dynamic workloads. © 2015 IEEE.


Published in:
2015 IEEE International Parallel and Distributed Processing Symposium, 904-913
Presented at:
2015 IEEE International Parallel and Distributed Processing Symposium (IPDPS), Hyderabad, May 25-29, 2015
Year:
2015
Publisher:
IEEE
Laboratories:




 Record created 2015-11-23, last modified 2018-06-22

Preprint:
Download fulltext
PDF

Rate this document:

Rate this document:
1
2
3
 
(Not yet reviewed)