Data Management in the Cloud

Speaking at a joint meeting of the SDForum Business Intelligence and Cloud Services SIGs on July 20, 2010 at the SAP facilities in Palo Alto, Daniel Graham of Teradata explained that cloud computing is now good enough in performance, reliability, and security for many, though not all, applications. He encouraged the audience to get into the cloud, try it out with a development project, gain experience, learn the bumps, so as to be ready when the boss demands that you move to the cloud in order to reap huge cost savings.

To clarify the marketing hype, Graham defined cloud computing as having five key attributes: on-demand self-service, broad network access, resource pooling, rapid elasticity, and measured services. Emphasizing the importance of cloud computing, Graham quoted a prediction by Gartner that by 2015 twenty percent of computing equipment will be located in the public cloud. Organizations adopt cloud computing, according to Graham, for cost savings and agility, with agility being the greatest business value gained through the cloud.

Addressing the key question of security, Graham said that the majority of security concerns around cloud computing are unfounded. He asked whether anybody in the audience could guarantee that no hacker would ever breach their corporate firewall. Seeing no hands go up, Graham said that organizations do not talk about breaches but that does not mean that breaches do not occur. In contrast, cloud vendors must guarantee security because any breach would put them out of business. Graham noted that the federal government, even in the area of defense, has become a leader in cloud computing. He argued that organizations can protect their data by encrypting data transfers and encrypting databases stored in the cloud.

Graham pointed out that cloud computing differs from on-premise computing. In the typical scenario, your virtual machine runs on commodity hardware shared by virtual machines rented by other organizations. Your virtual neighbors on a physical machine might run code that misbehaves and hogs resources, meaning that the performance of your applications may vary. For a mission critical application, your organization might find this variation unacceptable. And for now, according to Graham, organizations are keeping their big machines and mission critical applications in the datacenter. But performance ranges in the cloud, Graham said, are good enough for most application; not every application needs the ultimate in performance and reliability.

Graham told the audience that one or all of the traditional BI components may move to the cloud: the OLTP data source, the ETL (extract, transform, load), the data mart, and the BI tools and applications. What BI practitioners must now think about is data transfer, which has an explicit cost in the cloud. Graham recommends compression and, wherever possible, colocation, so that data does not need to be transferred over the Internet. Speaking of data that originates in cloud applications, Graham recommends that “if data starts in the cloud, leave it there.”

Because of customer interest in cloud computing, Graham told the audience that Teradata has worked with cloud vendors to make its BI software available in the cloud for small to medium-sized BI applications. Running Teradata on commodity hardware provided by other vendors represents a first for Teradata, which traditionally bundles its software with its own hardware for optimal performance. And Graham still recommends the on-premise deployment of Teradata hardware and software for large data warehouses. But for those customers who want to move to the cloud, Teradata will work with them to meet their business needs.

Leave a Reply

Please log in using one of these methods to post your comment: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: