Moving from physical servers to VMware virtual machines often creates new backup and storage challenges, as law firm Hooper Lundy & Bookman, P.C. discovered when it consolidated about 50 servers across offices in four states. The company solved its issues with a scale-out NAS system that uses no LUNs, volumes or RAID.
The law firm, which represents healthcare providers and suppliers, opted for Exablox's object-storage OneBlox appliance over traditional disk backup alternatives, such as EMC's Data Domain. HLB uses OneBlox in combination with Veeam Software's Backup & Replication application for virtual machine backup and disaster recovery for offices in San Francisco, Los Angeles, San Diego and Washington, D.C.
"We did a migration from physical servers to VMware virtual machines over the last couple of years," said Greg Williams, chief technology officer at HLB. "We had four Hewlett-Packard tape libraries and we were using CA ARCServe backup agents on each computer. We quickly found out that does not work in a virtual environment. Almost all virtual backup systems entail disk-to-disk systems."
Initially, HLB repurposed three physical HP servers with on-board storage that became the firm's Veeam repository, while keeping some tape. However, the storage on the HP servers didn't scale to the firm's needs.
"That was our disk farm before Exablox," Williams said. "But there were two issues. We run out of space and that directly impacts how many copies you have. There are only so many disks you can stuff into an HP server. Also, we are not protected in case of a fire because the disk farm is in the same building as the data you are protecting.
"You are in stellar shape for file retrieval, but not protected from an earthquake."
A 2U OneBlox device has eight drives and can handle a mix of SATA, SAS or solid-state drives (SSDs). The architecture uses a ring concept, which is a number of OneBlox appliances or nodes that present a single global file system.
"Before I would have had to buy a refrigerator and I'd waste huge amounts of money buying this behemoth," Williams said. "Exablox gets you out the problem of, 'What if I buy the wrong size?' I buy big enough and then add as I need. Plus, it takes different drive technologies, so you have a lot of control over what drive you want. Guys like me like that."
Exablox's deduplication capability helped solve a problem with Veeam backups: It consumes a lot of storage.
"Veeam copies files, but it brings up an entire terabyte. It's a full backup," Williams said. "It does dedupe within a job, but not against a file of last week. It dedupes on a per-job basis, but not against old data."
The system also met HLB's off-site replication need. Williams said the firm still uses its HP server disk farm for faster local reads with OneBlox serving as a long-term archive. The VMware servers connect to the disk farm via a DataCore storage area network (SAN) over Fibre Channel, while the HP servers sit in front of the Exablox system.
"Veeam makes copies of data onto the HP local disk, and another copy onto Exablox," Williams said. "If I need a current restore, I get it from the HP servers since it will be the fastest. If I want a backup from a month ago, it is on Exablox."
The system's use of object storage rather than RAID was also attractive to Williams. He said as you add more Exablox systems, they spread copies of data across appliances.
"You are building a very reliable system that can suffer drive failures. Performance is miserable when you are doing a RAID rebuild," he said. "Now as drive sizes have gotten to four terabytes, that RAID rebuild that once took a day, now takes a week. That's only going to get worse as drive sizes get bigger."