University of Leicester dumps SANs and LUNs for Cloudian object storage

The University of Leicester has ditched its Dell-based SAN backup targets and moved to Cloudian object storage in a move that has seen it cut reliance on Fibre Channel fabrics and LUN storage for Commvault backups.

The migration to Cloudian has also allowed the university to cut rack space used by 50% – from 48U down to 24U – and freed it up to potentially move backup storage to remote datacentres or the cloud.

Leicester has more than 20,000 students and 4,000 faculty and staff. Systems specialist Mark Penny and his team provide backup for all home directories, corporate systems and research data (including HPC data) in a mixed environment encompassing Windows, VMware and Lustre.

The existing Dell-based SAN storage – comprising end-of-life PowerVault and Compellent arrays – connected to 10 media servers was showing its limitations, most notably in the tight dependencies between LUNs and media servers that had resulted in major issues.

“We’d had the same basic design since 2010,” said Penny. “Because each LUN was tied to a media server, if we lost a media server then we’d lose access to the LUN. That meant you couldn’t restore that LUN and, if we lost the index, we couldn’t do a backup either.”

Penny said that had happened, for example, when several LUNs were lost following Windows patching. “That really highlighted things,” he added.

Penny’s team had investigated Cloudian object storage for a previous project, and when the time came to upgrade the storage infrastructure, the systems specialist was attracted to it again.

“With object storage S3, the media agents could become stateless,” said Penny. That is, the rigid one-to-one mapping between LUNs and media servers would be gone. So, if a media server became unavailable everything else would continue to work.

“If you lost a media server you couldn’t get to anything – it’s how Fibre Channel works – and you’ve lost that server,” he said.

Penny carried out a proof-of-concept and later deployed 12 (later adding three more) Cloudian nodes on HPE Apollo 4200 server hardware, with 24 12TB drives on each.

Cloudian’s core product is object storage based on the Apache Cassandra open source distributed database. It can come as storage software to be deployed on commodity hardware, in cloud instances or in hardware appliance form. Its Hyperfile file access – which is Posix/Windows-compliant – can also be deployed on-premise and in the cloud to provide file access.

Daily backups (incremental) amount to between 5TB and 25TB. Total capacity on the Cloudian storage is now up to 2.9PB useable (4.3PB raw). If you add new nodes, it is simply a case of deploying them and doing a re-balance of the data, although that can take some time.

Instead of Fibre Channel, connectivity is now via 40Gbps Ethernet. Under the new configuration Commvault takes backup “chunks” and writes them to S3 storage. So, what are the benefits for the university?

“A lot of it is around the flexibility it has given us and what we can do with backups and Commvault,” said Penny “We can now potentially move the backup storage infrastructure off-site. With Fibre Channel that wasn’t possible because it won’t stand the latency. You get timeouts and CRC [cyclic redundancy checking] errors.

“We may possibly use the Janet datacentres in Leeds or Slough, which we can now because the system will accept latency,” he added.