The new mirrored setup uses SIOS host-based replication to copy data in near real time between the primary and disaster recovery sites.
The university has a research staff of 2,000, with data that ranges from CT scan images to Word documents, but historically this was in scattered silos of storage that were difficult to manage and vulnerable in data protection and disaster recovery terms.
“We had islands of storage that researchers had created and it was all based on a Novell file services architecture, which meant we couldn’t provide multi-terabyte pools,” said Mark Griffin, technical architect at Nottingham university.
In 2010, the university began to evaluate shared storage alternatives that could deal with storage volumes that would range from a few gigabytes to tens of terabytes, were reliable and would meet disaster recovery requirements.
Following an evaluation process the university implemented around 0.5PB of storage across two datacentres in Nottingham.
Access to the research data is front-ended by a pair of HP X3800 Network Storage Gateway NAS heads running Windows Storage Server with Fibre Channel connectivity to two 45TB HP P2000 SANs running HP’s Midline (ie, nearline) SAS drives. These are 7,200rpm drives that have a 6GBps throughput and are suitable for non-transactional workloads such as file access.
Download additional resources on disaster recovery
In addition, Griffin’s team deployed three more expansion disk enclosures with 150TB each of capacity. The whole mirrored configuration uses SIOS SteelEye DataKeeper Cluster Edition – recommended by reseller Phoenix – which is a host-based replication product that provides asynchronous replication over a 10Gbps link and failover and failback between the sites from within the Windows Storage Server nodes.
Host-based replication is a software-only solution that is installed on servers from where it intercepts data traffic and replicates it. It can be affected by other application workloads, but has the advantage of being storage agnostic where most array-based replication products will only replicate to the same supplier’s arrays.
Griffin said the chief advantages of the setup are that it allows for smooth patching procedures and viable data protection and disaster recovery provision.
“We need to be able to failover and failback rapidly between datacentres. We have a regular patching cycle and we wanted to be able to patch the server, fail it over, reboot the server then failback,” he said.
Griffin opted for the SIOS host-based replication over the array-based replication available in the HP SAN products because of shortcomings with array-based replication.
“SIOS is low cost for what it does. It allows us to failover and failback with no issues, and if we wanted to move from HP to EMC storage it would still work. Also, it’s not array-based. Our experience with hardware replication has not been as simple as you would expect. Array-based replication is good for failover, but failback is no good,” said Griffin.
The big picture is to provide adequate disaster recovery provision for the university’s research data. “We can never lose the crown jewels of the university,” he said.