One talk by University of Massachusetts researcher Emmanuel Cecchet centered on Dolly, a system focused on dynamic provisioning of database resources in the cloud.
Cloud-based deployments work particularly well for Web applications, since the dynamic resource provisioning and metered pricing cloud platforms provide are ideal for handling the typical peaks and valleys of usage they receive, Cecchet and other researchers note in a paper on Dolly, which was co-authored by Rahul Singh, Upendra Sharma and Prashant Shenoy.
However, dynamic provisioning so far has focused mostly on the front end or middle application server tier, they argue.
Over-provisioning databases in anticipation of a peak demand instead of adding and subtracting capacity dynamically stops applications from fully taking advantage of cloud services’ metered pricing, they add.
Dolly tries to solve the problem through a system that uses virtualization to create database replicas, as well as a provisioning algorithm that allows administrators to squeeze the most efficiency out of their cloud resources.
The researchers have implemented a prototype of Dolly and plan to release it as open-source software.
Also presenting Friday was James Starkey, founder and CTO of the startup NimbusDB.
Starkey, who is also the creator of InterBase and a former senior software architect at MySQL, noted the complex, varying demands cloud computing places on databases, such as a 10,000-strong deployment of mobile collaboration applications.
All the mobile clients need a back-end database running around the clock, but in terms of usage, “some will go nowhere, many will have modest requirements and some will go off the chart,” he said.
NimbusDB, which Starkey began developing a few years ago, is an elastic, SQL-based relational database that adheres to the ACID (atomicity, consistency, isolation and durability) concepts for database transaction integrity.
The database coins some new terminology as well. A “chorus” refers to the group of nodes that implement the database, while the database is composed of objects called “atoms,” according to a NimbusDB presentation.
Nodes can be added or subtracted from the live system as needed.
NimbusDB’s website provides a few indications of the product’s goals. “Imagine telling your database system to start using some spare machines in London and having the extra capacity seconds later,” it states. “Imagine having your database partly in your datacenter and partly on a public cloud, or ‘cloudbursting’ peak workloads to other clouds.”
Most of the components of NimbusDB have been implemented, Starkey said, but he would not provide a release date.
