Britannia is the second largest building society in the UK, with 5,000 employees at sites across the UK. The institution offers a range of financial services, including mortgages and other loans, credit cards, savings accounts and insurance. It runs financial, mortgage, investment, CRM and procurement applications on 130 Unix and 100 Windows servers.
So Britannia implemented Data Domain DD580 disk arrays, with data deduplication at each Leek site, which sent data on to Bristol as part of a wider project to implement business continuity provision and set up a secondary data centre.
"Primarily we were driven to look at dedupe by storage capacity issues," Mathias said. "Data volumes were getting bigger with regulatory requirements, and we wanted to minimise that. Data deduplication seemed the most sensible thing to do, especially now that CPUs are up to the task of crunching the data as it goes to the device."
Each of the two Leek data centres backs up to its Data Domain unit locally, from where it is replicated to Bristol. Here the local DD580 is able to further deduplicate the data by comparing data and to eliminate redundancy from the two Staffordshire sites.
Dedupe ratios achieved vary between 10:1 and 20:1 in a typical week, says Mathias, depending on the nature and volume of changes during that week. "Last week's ratio was 18:1," he said. "But under some conditions, we have seen individual servers compress as high as 40:1."
The first phase of the project -- replicating to offsite backup with the new technology -- has been completed. During the second phase, disk will completely replace tape as the primary target for all backups by year-end.
So far, the key benefit for Britannia has been the elimination of tape at the Leek data centres. For Mathias, the benefits of backing up over the wire means no headaches from dealing with removable media, no reliance on third parties and lower management costs.
"Deduplication has saved staff admin time massively," Mathias said. "There is no opportunity for media loss, no forgetting to change media. The big benefit comes when comparing to maintenance costs for traditional tape silos, which cost as much to keep going in one year as we'd spend on one Data Domain device. It's much more cost-effective, and there are no ongoing media costs."
Mathias declined to name the other data deduplication vendors he looked at during the procurement process, but he did say that the main differentiator was whether the product used inline or post-process deduplication.
Inline data deduplication checks for redundancy in data as it is streamed to the device and replaces duplicate data with pointers to the original instance as it passes through. Post-process deduplication receives all the data in a job, then processes it for redundancy.
Inline processing can often take longer as checks are made on data flows as they move to the device, but post-process methods require more disk space, whichwas what put Mathias off. "Our main issue with the majority of [deduplicating products] was them being post-process, which needs holding disk space as well as disk space on which to put the deduped data."
Mathias said he was interested in using the Data Domain products as NAS targets, but they weren't designed to operate that way at the time Britannia looked at them. "We already use it as a backup target,' he said. "But we'd like to investigate whether it can be used in future as storage for Windows data."