An Elastic Virtualised Infrastructure for Data Services - Han Li UNSW
Many enterprises are now adopting cloud storage systems (e.g. key-value stores) for data provisioning services on servers leased from Infrastructure-as-a-Service (IaaS) cloud providers. One of the advantages of cloud computing is rapid elasticity. That is, the ability to deal with increasing demand by acquiring new virtual machines, (i.e., scale up) and to shut down the machines in the face of declining demand (i.e., scale down), to save on economic costs. However, state-of-the-art cloud storage systems are unable to efficiently scale up and down in a timely manner.
My research focus is on how to design and implement a key-value store that is elastic to the workload demand. The incentives of this research are: i) to improve the system's performance by scaling up to deal with increased workloads; ii) to minimise the on operating costs by scaling down when the workloads decrease. This research is tied to the Services Aggregation project within the CRC, and also involves with working with SIRCA for supporting on-demand news article processing. The outcomes of this research will be beneficial to the companies that leverage IaaS cloud to serve large volumes of data with dynamic workload.
Specifically, I have designed and implemented ElasCass, a shared-nothing, key-value store that builds on Apache Cassandra. ElasCass enables rapid elasticity, by using a novel online data partitioning scheme for lightweight data migration, and a novel data placement algorithm that ensures high data durability for a dynamically scaling system. This work has been published in Middleware 2013, a very selective system conference. Smart Services CRC has also filed a provisional patent application for this research effort.
I have now come to the stage of writing my final thesis. In the next six months, I will be looking for job opportunities in the area of cloud data management. Please do not hesitate to contact me via [email protected], or talk to me in the CRC Annual Conference.