Skip to content (access key 's')
Logo of Technion
Logo of CS Department
Logo of CS4People

The Taub Faculty of Computer Science Events and Talks

ceClub: Reliablity and Efficiency in Cloud Storage
event speaker icon
Ittay Eyal (EE, Technion)
event date icon
Wednesday, 13.02.2013, 11:30
event location icon
EE Meyer Building 861
Advances in data center technologies have led to extensive use of data centers to store large volumes of data in a managed distributed system. The users of such systems have increasing expectations of both efficiency and reliability.

Recent years have shown that even the largest cloud storage providers occasionally fail, and users have to replicate data among multiple providers to obtain reliability. However, classical replication techniques (e.g., ABD) are not applicable here, since storage services typically export only a key-value store (KVS) interface, with functions for storing and retrieving values associated with unique keys. In the first part of this talk, we present the first efficient, wait-free algorithm that emulates multi-reader, multi-writer reliable storage from a set of potentially faulty KVS replicas in an asynchronous environment. We implemented the algorithm and tested it using real-world providers, and we demonstrate through simulation its low overhead in terms of space and speed in various scenarios.

In the second part of the talk we introduce ACID-Rain: ACID transactions in a Resilient Archive with Independent Nodes. ACID-Rain is a novel architecture for efficiently implementing transactions in a distributed data store - a sought-after but difficult to achieve capability. ACID-Rain uses logs in a novel way, limiting reliability to a single tier of the system. A set of independent nodes form an outer layer that caches the data, backed by a set of independent reliable log services. ACID-Rain avoids collisions between transactions by using prediction to order the transactions before they take actions that would lead to an abort. ACID-Rain is efficient in overcoming failures, and its throughput increases linearly with the number of nodes.

Ph.D. Seminar, under the supervision of Prof. Raphi Rom and Prof. Idit Keidar.

Bio: Ittay Eyal is a Ph.D. candidate in the department of Electrical Engineering under the supervision of Prof. Raphi Rom and Prof. Idit Keidar. Ittay Received his B.Sc. from the Technion in 2007, and did internships with IBM, Yahoo! and Cornell university. His main research interests are distributed storage and reliability in large scale systems, and robust aggregation in sensor networks.