DBHAaaS – Database High Availability As A Service For Cloud Computing
Metadata[+] Show full item record
On conventional database systems, the recovery manager applies transaction Undo or Redo operation or a combination of them to recover the last consistent state of the database from a system failure. Transaction redo, compared to undo, helps to shorten the system downtime so the execution of transactions is managed in such a way that majority of transactions require redo to recover the database. To further reduce the recovery time, the recovery module uses “Checkpoint” operation. Even though it is possible to eliminate transaction redo altogether, the conventional system architecture, however, is not capable to deploy innovative approaches. The availability of “Virtual” machines on cloud has given us an architecture that makes it possible to completely do away with transaction redo which allows us to eliminate the effect of system or transaction failure by taking the database to the next consistent state. In this dissertation, we present a novel scheme of eliminating the effect of such failures by applying transaction “roll-forward.” Our approach intelligently applies roll-forward from the point of failure which removes the effect of system failure on the database. We refer to our system as AAP (Always Ahead Processing). In AAP a transaction always executes to completion. We have made forward execution of transactions persistent by combining together transaction execution, transaction failure and its subsequent recovery as one seamless operation. Unlike legacy recovery scheme, in our approach transactions roll forward from the point of failure while concurrently executing other transactions. As a result, system downtime during recovery is eliminated. The end result is a database system with high availability and fault-tolerance. Our work enables cloud providers to offer transactional HA-DBMS (Highly Available – DataBase Management System) as an option that too with multiple data sources not necessarily only relational. AAP is independent of the underlying cloud architecture and therefore can be used in different type of cloud settings like public, private, hybrid or federated. In a federated cloud, the location of the physical host machine(s) is very important for fast and responsive HA (High Availability) service. We designed and developed a tool that will aid AAP to find the nearest physical host. We call this tool IGOD (Identification of Geolocation of clOud Datacenter). Apart from aiding AAP in a federated cloud, IGOD is an independent tool and can also be used for enforcing privacy and security in cloud datacenters; in particular for HIPAA compliant data storage. Our prototype demonstrates AAP’s HA, fault tolerance and the elimination of system downtime during recovery from a failure.
Table of Contents
Introduction -- Literature review -- DBHAaaS - AAP -- Identification of Geolocation of Cloud Datacenter -- Conclusion and future work