Production VMs tend to drift from their desired (standardized) state, creating compliance, maintenance, and governance problems. On a cloud scale, these problems can amplify rapidly, increasing operational risk and cost. Current agent-based approaches to address this problem don't work very well in practice. Customers view agents as disruptive; operating in the production environment constrains what the agent can do; and the agent is itself susceptible to accidental or malicious disruption.

Origami is a system that offers a radically different solution. It periodically checkpoints the (disk and memory) state of running VMs using the hypervisor's built-in snapshot capability, and processes the checkpoint images offline using data analytics algorithms. A key advantage of this technique is that it is agent-less and therefore non-disruptive to the customer's production environment. By increasing the checkpointing frequency and scaling the analytics algorithms to operate on streaming image data, Origami is able to deliver near real-time drift detection services.

In essence, Origami transforms an online systems management problem over VMs into an offline data analytics problem over images. This has game-changing implications that extend beyond the realm of conventional systems management. An example of this is the knowledge base created by Origami, of recurring software patterns discovered through feature extraction, with annotations about their behavior (e.g. propensity for defects). Such an automatically curated knowledge base is valuable even outside the production environment in which it was constructed.

Origami technology is already being integrated into a Tivoli product, is being evaluated by the GTS cloud migration team, and has received positive customer feedback. This talk will describe the image analytics technology behind Origami, and dmonstrate a prototype of the system.

Vasanth (Vas) Bala leads the Virtualization Runtime and Tools department at IBM Research. His research interests are in managing software systems as data, by operating on VM images, instead of VM instances. He also leads the virtualization strategic initiative for the worldwide IBM Research division, where he is responsible for overseeing IBM's technology strategy in the virtualization space. Vas has been previously employed at HP Labs, and was founder of Liquid Machines Inc, a startup in the Boston area that was acquired by Check Point Software.