You are here

NSF-funded research aims to give cloud users insight into hardware resource usage

Professor Yinqian Zhang will develop much-needed monitoring tools to enable cloud computing customers to efficiently allocate computing resources.

Interacting with the cloud has become part of modern life, with online email services, document storage, shopping and more powered by cloud computing. Businesses also increasingly rely on the cloud—global spending on cloud services is projected to surpass $305 billion in 2018, according to Garner.

With its massive pooling of computing resources, cloud computing can offer organizations of all sizes the prospect of lower IT costs, fewer administrative headaches and rapid scalability. But sharing physical machines and critical computing resources—like processor, memory and storage—among multiple users means performance can be negatively impacted by neighboring activities. Currently, cloud users have no way to monitor hardware resource usage to see if their applications and computations might be affected by other activities.

Prof. Yinqian Zhang and his team are developing monitoring tools to enable cloud computing customers to efficiently allocate computing resources.Funded by a three-year $500,000 National Science Foundation grant, Assistant Professor Yinqian Zhang is working to develop novel techniques that will enable cloud users to monitor resource usage of physical cloud servers without assistance from cloud providers. He holds a joint appointment at The Ohio State University in both computer science and engineering as well as electrical and computer engineering.

The current lack of monitoring ability and performance guarantees can mean delays and slow loading times for users of customer-facing services. That affects cloud users’ willingness to fully embrace the economic benefit of cloud computing, Zhang explained, especially by researchers and others whose applications require stable and predictable runtime environments.

“This is part of the reason why people don’t run time-sensitive scientific computations in the public cloud,” he said. “The environment is not controlled and you don’t know what’s really going on.”

Building on his expertise in cloud security research, Zhang plans to transform techniques traditionally used for malicious purposes into tools that will help cloud users gain insight into critical resource usage.

Specifically, researchers plan to use self-monitoring virtual machines, which leverage nested virtualization technology and side-channel analysis techniques, to give users a tool that will enable them to see how resources are being utilized so they can allocate them more efficiently.

“The idea here is rather than using this technique to attack others, we want to use it to do something good—gain visibility into the hardware layer and periodically sample what is really going on in the server. Are there other virtual machines? Do they consume a lot of memory? Do they consume a lot of cache or narrow bandwidth?” Zhang said. “If we observe something like that, that means the performance of my computation will degrade.”

New educational tools will also be developed, including Amazon machine images which enable a derivative cloud to be created on top of public clouds. Using this derivative cloud, students can easily obtain hands-on experience with cloud computing.

“If I’m teaching a course and want students to use a cloud service now, it’s difficult,” Zhang said. “But if I can run a derivative cloud on top of the public cloud, that means I can operate a small-scale cloud environment to allow students to explore cloud computing and manage their own account.”

The researchers will also create open-source tools that enable self-monitoring virtual machines in public clouds and will freely share the source code and Amazon machine images to encourage adaptation and adoption of their techniques.

by Candi Clevenger, College of Engineering Communications,