Financial Giant Uses Deduplication to Reduce Virtual Storage CostsBy Drew Robb
June 23, 2010
SunGard Financial Services was experiencing backup woes in its highly virtualized environment. Using tape along with VMware Consolidated Backup (VCB) and Symantec Backup Exec was proving problematic.
"We didn't have enough tape bandwidth to complete large backups and were experiencing high rates of backup failure," said Guy Chapman, a senior engineer for storage and virtual infrastructure at SunGard Financial Services, who is based in London.
The company looked at various options as an alternative to tape, including storing data on SATA disks within a SAN. Eventually, however, it switched to EMC Data Domain for deduplication and Symantec NetBackup software. Some backup remains on tape, but much of it has switched to the Data Domain appliance.
"Data Domain-based deduplication worked out at about one fifth of the cost of SATA/SAN storage," said Chapman. "We have also reduced our backup and recovery costs per VM [virtual machine] by 10 times."
SunGard Financial Services is one of the four business units of SunGard. While SunGard Availability Services is perhaps better known due to its disaster recovery services, SunGard Financial Services is actually a larger revenue generator. It contributes more than 50 percent of the annual $5 billion-plus of SunGard by delivering software and IT services to institutions in virtually every segment of the financial services industry.
Based in London, but with global operations, the data center of SunGard Financial Services uses EMC Clariion and Celerra arrays in its SAN. It currently has 1,500 virtual machines (VMs) distributed across eight sites. Most of these VMs operate in a development and test environment. When it came to backing them, the inadequacies of VCB meant that priorities had to be set. Those VMs operating in a mission-critical production setting were always backed up. Those in development and testing, however, were not. Chapman describes a "kit of parts" being needed to build VCB backup. Despite a lot of attention, backup jobs would fail.
This situation was exacerbated by a distributed infrastructure. Many smaller sites existed throughout London, and each business unit ran its own backup and DR plan. The company engaged in a metropolitan consolidation project to get rid of most of the small sites, focus IT operations at a large data center at Canary Wharf in London, and centralize backup.
"We needed to provide a reliable and consistent protection strategy for our virtual and physical real estate," said Chapman. "Our three tape silos were unable to complete backups on time."
With more than 190 TB of data at the front end, SunGard Financial Services looked at how to back all this up and replicate it to a recovery site without breaking the bank. That led the company to deduplication as a means of reducing the volume of data that needed to be stored.
Dedupe works by a) preventing duplicate files from being backed up and b) storing only one copy of everything rather than conducting a full backup on a daily or weekly basis. The two final dedpe candidates were EMC Data Domain and Symantec PureDisk. The latter was priced attractively if purchased with Symantec NetBackup. However, Chapman said it has relatively high associated hardware costs -- five media servers -- as well as power and cooling demands, and the need for more back-end storage to support it.
"Data Domain met our business requirements and could easily be implemented into our environment," he said. "The reference calls we made were positive and the technical presales contacts were excellent."
In terms of results, the company has been able to squeeze 190 TB at the front end to only 14 TB on a Data Domain DD690 appliance using NetBackup software. New incoming data achieves compression rates of 20 to one.
Tape, though, continues to be used as a long-term archive. It has been eliminated for offsite backup. Instead, the company placed another DD690 at an offsite facility and replicates its data to it.
"We now have fast reliable recovery of replicated data," said Chapman.
Another benefit is nearline storage of VMs. Many VMs in test and development are used for only a short while, yet they can take a long time to configure. While some might never be needed again, it is smart to keep them around just in case. But this requires a lot of storage. SunGard now puts them on the Data Domain box and can bring them back into use rapidly without clogging up primary storage with unused VMs. This works out at about one tenth the cost of storing those VMs on primary disk.
"We have full confidence in the backup and security of or virtual infrastructure," said Chapman. "The time savings of implementing deduplication are equivalent to one full-time employee due to the elimination of so much backup administration and troubleshooting."
Drew Robb is a freelance writer specializing in technology and engineering. Currently living in California, he is originally from Scotland, where he received a degree in geology and geography from the University of Strathclyde. He is the author of Server Disk Management in a Windows Environment (CRC Press).