Enterprise Cloud Bursts – Innovation of Stratospheric Proportions

A new technology developed by researchers at the University of Massachusetts Amherst and IBM is taking the Enterprise Cloud to new heights. Codenamed Seagull, the cloud management software is capable of improving cloud performance during peak usage “bursts”, when a large number of users simultaneously access cloud applications.

In essence, cloud bursting is a technique used by network administrators to save on data center costs. While public cloud hosting can often be comparable in price to private data center hosting, using a combination of both can lead to significant cost savings. For instance, if a certain online store requires 5 processors most of the time, and the load increases to 10 processors on Saturdays and Sundays, private data centers would often provision the full machine with 10 processors for the entire week. A hybrid cloud, on the other hand, would enable the system administrator to use the 5-processor machine throughout the week, and only transition to a 10-processor machine in the public cloud on weekends to accommodate the high volume of traffic.

The Seagull cloud management software attempts to make the process of “bursting” from the private cloud to the public cloud more streamlined. Most current cloud-bursting tools request the system administrator to manually decide when the switch apps to the cloud. The Seagull software, on the other hand, uses integer linear processing (ILP) algorithms such as CPLEX and a custom greedy algorithm to automatically decide when the move the virtual machines, and how to provision them for optimal performance and cost savings.

The most powerful component of Seagull, however, is the intelligent precopying algorithm. Instead of requiring the entire virtual machine to be transitioned to the cloud when performing a burst (an operation that could take several hours to several days, depending on the disk size), Seagull instead predicts which virtual machines will require the cloud, and “precopies” the images onto the public cloud. This technique ensures that the entire disk will not have to be synchronized when the time comes to burst. Instead, the server will only need to replicate recent changes, and the transition time is often reduced from several hours to a few minutes.

Seagull is predicted to reduce data center operating costs during cloud bursts by 45%, primarily due to the precopying algorithm. Since data transfer costs are often high compared to simple storage costs, replicating changes instead of copying the entire server image is predicted to significantly reduce costs, especially on servers with frequent bursts.

What’s missing in the paper, however, is a justification of the optimization algorithm. While the paper assumes that system administrators are unintelligent and make bad decisions regarding how and when to perform cloud bursts, the paper does not actually show the algorithm’s real-world performance characteristics. Instead, a blanket statement of “we expect a good workload predictor will have less than 10% error,” is used to gloss over actual proof of the concept.

In addition, one of the primary sources of the performance improvement is not in the bursting itself, but in local private inter-machine optimization. The comparison algorithm does not optimize locally, and only transitions virtual machines from private to public cloud. In reality, the majority of the system administrator’s time would be spent optimizing the local cloud, as opposed to focusing on the bursts themselves.

The paper does show significant promise though, and portrays an exciting future for automated private/public cloud optimization and deployment. A more rigorous analysis of real-world performance would help prove the concept, and perhaps instead of demonizing system administrators, the paper’s authors could work together with them. By converting Seagull into an intelligent recommendation and deployment tool, the software could provides insights for sysadmins and alert them to possible inefficiencies in the cloud. In this way the system administrator could still make the final decision based on practical experience and intuition, yet leverage Seagull for the optimization recommendations and precopying technology.

Written by Andrew Palczewski

About the Author
Andrew Palczewski is CEO of apHarmony, a Chicago software development company. He holds a Master's degree in Computer Engineering from the University of Illinois at Urbana-Champaign and has over ten years' experience in managing development of software projects.
Google+