Gayana Chandrasekara
Adaptive Quorums for Cloud Storage Systems
Tese submetida para provas de mestrado em Engenharia Informática e de Computadores
Instituto Superior Técnico, Universidade de Lisboa.
Abstract
Cloud storage systems rely on replication for reliability. Typically,
each data object is stored in multiple nodes to ensure that data
remains available in face of node or network faults. Quorum systems
are a practical way to ensure that clients observe consistent data
even if some of the replicas are slower or unavailable. Previous work
has shown that the performance of a quorum based storage system can
vary greatly depending on the workload characterisation and that
significant gains can be achieved by carefully selecting the size of
write and read quorums. In this work we are interested in multi-tenant
storage systems, that are faced to heterogeneous works. In these
systems, for optimal performance, different quorums may need to be
applied to different data. Unfortunately, keeping different quorum
systems for different objects significantly increases the amount of
metadata that the storage system needs to manage. The challenge is to
find suitable tradeoffs among the size of the metadata and the
performance of the resulting system. The thesis explores a strategy
that consists in identifying which tenants and/or objects are the
major sources of bottlenecks in the storage system and then performing
fine-grain optimization for just those objects, while treating the
rest in bulk. We have implemented a prototype of our system and
assessed the merits of the approach experimentally.
Publicações
- Adaptive Quorums for Cloud Storage Systems
- Gayana Chandrasekara
- MSc Thesis. Instituto Superior Técnico,
Universidade de Lisboa.
- July, 2015.
- Available BibTeX, MSC Thesis, and extended abstract.
- Q-OPT: Self-tuning Quorum System for Strongly
Consistent Software Defined Storage
- M. Couceiro,
G. Chandrasekara, M. Bravo, M. Hiltunen, P. Romano,
L. Rodrigues.
- In Proceedings of the 16th ACM/IFIP/USENIX
Middleware conference, Vancouver, Canada, December 2015.
Luís Rodrigues