It begins with an IT administrator putting a virtual disk on one of the available storage devices and going home for the night. It percolates overnight and by the time the admin is back to work, it explodes, sending flack in the form of 50 or so messages from angry users who cannot access their environments because the storage array the virtual disk was delegated to did not have enough room.

After hours of troubleshooting the system, the admin finally finds the cause of the issue and shifts the excess load to a different datastore using VMware's Storage vMotion. According to VMware Staff Engineer Irfan Ahmad, this is a common occurrence in the lives of IT admins that run virtualized environments. 

VMware's R&D department has been hard at work to prevent such an issue from happening. The solution they are working on has not been turned into a product and there are no concrete plans at the moment as to what shape it may take if does become a product in the future. "This is just some cool stuff we've been playing with in the lab," Ahmad said during a presentation at VMworld 2010, the virtualization giant's annual conference that took place in San Francisco this week. 

The technology, whose working title is "Storage DRS" introduces a new abstraction ÔÇô datastore groups ÔÇô said Minwen Ji, senior staff engineer at VMware. Instead of adding a virtual disk directly to a storage device, Storage DRS lets the admin add the disk to a datastore group, which includes several arrays, and automatically allocates resources to support it, making sure space and i/o load are balanced across all arrays. 

The company already has VMware DRS (Distributed Resource Scheduler), which automates load-balancing of CPU and memory resources across a cluster of servers using vMotion. It also has Storage I/O control (SIOC), which automates prioritization of VM i/o's across servers, fixes the "noisy neighbor" problem and provides proportional sharing of shared i/o resources. 

"We have DRS, we have storage i/o control, but there's still an elephant in the room," Ahmad said, referring to the lack of automated i/o and load management for resources supporting VM storage. 

Ahmad said he was confident no-one else was working on this problem and VMware researchers have already spent about three years on it. "We studied published material, we studied academic literature, we studied state-of-the-art products that people have developed," he said.