University clusters Scale Computing HC3 boxes to eliminate sprawl

The "desktop support guy" at Auburn University learns about storage by implementing a Scale Computing HC3 cluster to eliminate storage silos.

Auburn University's College of Architecture, Design and Construction is implementing hyper-converged storage with a Scale Computing HC3 system to consolidate storage capacity that had been spread among dozens of standalone servers.

The college is replacing most of its 40-plus Dell PowerEdge servers with direct-attached storage (DAS) with a four-node cluster of Scale Computing's high-capacity HC4000 platform. The Dell boxes sit in isolated IT silos scattered across various sub-departments.

The installation at the 159-year-old Alabama university involved creating virtual machines (VMs) to migrate 15 TB of storage from DAS controllers to the HC4000 cluster. Auburn spent approximately $90,000 to purchase Scale Computing's "Cadillac model in an attempt to future-proof" storage growth, said IT coordinator Joel Beckum.  

"It's probably way more capacity than I need right now, but because I'm unifying the storage and VMs, we needed extra storage. We're a design school, so we have a lot of large media files. I needed it more for the storage than anything else. My plan is to get us down to 15 to 20 physical servers at a maximum," Beckum said, including winnowing file servers down from nine to no more than two.

Storage management for the 'desktop guy'

Auburn has been running its HC4000 cluster in a test bed since November. Beckum said several PowerEdge servers will remain in use as a backup repository for VM metadata to reduce dependence on a legacy Dell PowerVault TL4000 automated tape library.

The Scale Computing HC3 system is designed using open source Linux Kernel-based Virtual Machine (KVM) as its core hypervisor. The systems are powered by the HyperCore automated management software, a bare-metal hypervisor that is written around the KVM software kernel.

Beckum describes himself as "desktop support guy" with limited technical knowledge of storage architecture. A year ago, he was thrust into a system administrator role and tasked with managing a dispersed server environment. One of his first challenges was getting a unified view into storage, both for performance monitoring and capacity planning.

"We have a lot of server sprawl," he said. "Before I got here, the previous practice in the department was to buy more Dell servers whenever we needed to add a function. It became unmanageable. I actually hired a student whose entire shift involves logging in to Windows Server machines to make sure they're patched and the backups are running."

Beckum reached out to Phil Forrest, the IT manager for Auburn's College of Math and Sciences, who suggested he explore hyper-converged systems that combine compute, networks, storage and virtualization in a single box.

Beckum said the two IT professionals selected Scale Computing for its storage management features. Because he lacked a secondary backup site, Beckum conducted failover tests by disconnecting one of the four HC4000 nodes on the VM cluster, then using the drag-and-drop interface to reconnect the node to the cluster after completing the migration.

"I was forcing a node to fail. The Scale Computing system took all VMs on the failed node and moved them to the live nodes, with no interaction on my part," Beckum said.

Auburn departments split cost of HC4000 nodes

A side benefit of HC4000, he said, is the ability to re-allocate Dell PowerEdge servers as a disaster recovery cluster to better manage backups. "The difference in visibility on our backups was like night and day," he said.

The College of Architecture, Design and Construction spurred a trend toward hyper-converged storage at Auburn. Four other colleges -- Math and Sciences, Forestry, Education and Human Services -- are dividing the cost of four additional HC4000 nodes to consolidate legacy storage and minimize overhead associated with VMware virtual desktop infrastructure.

"What sold us on Scale was their HTML5 Web GUI, which has a very shallow learning curve compared to VMware," Forrest said. "The ability to remotely administer storage from mobile devices is even more pronounced because we have such thin IT staff."

Next Steps

Scale Computing takes home gold in 2014 Data Storage Products of the Year

Scale Computing HC3 brings simplicity to school's virtualization

Dig Deeper on Hyper-Converged Vendors and Products