A Web 2.0 database architect and a construction-company chief technology officer say they've used embedded tools...
in Microsoft SQL Server 2008 and VMware's ESX Server to reduce the number of products in their shops for backup and disaster recovery, cutting costs and management headaches.
Glenn Berry, database architect for NewsGator, manages 25 TB of storage on 3PAR's InServe S400 high-end disk array. The majority of NewsGator's data is served through multiple instances of Microsoft's SQL Server 2008, which it recently put into production after beta testing.
NewsGator previously used a third-party tool, Quest SQL LiteSpeed, for SQL Server Backups, but its CPU utilization was high. "We weren't even using it anyway," Berry said. Having both data reduction and backup embedded within the application makes the server CPU more efficient, and reducing data at the host eases network congestion. With backup and application coming from the same company, NewsGator has one throat to choke and doesn't have to deal with finger-pointing.
For short-term restores, NewsGator also uses 3PAR's SAN snapshots to recover from errors and accidental deletions within an hour or less. If the company has to take down the SQL server for maintenance, data is mirrored to a legacy EMC CX-500 array kept around for that purpose. "I don't see the need to use third-party traditional backup tools at all," Berry said.
Disaster recovery with VMware
Ray Hayes, chief technology officer at J.F. Shea Co., said his company is happy with the Symantec Backup Exec software and Overland Storage hardware it's running for backups, such as several REO 9000 and REO 4000 VTLs, and NEO 2000 and ARCvault tape libraries at multiple locations.
But when it comes to replicating data among those locations for disaster recovery, the company is using a combination of VMware ESX Server stretched clusters and Citrix's Access Gateway secure VPN, rather than a third-party replication tool.
"The homebuilding business is so restricted right now" because of the mortgage crisis, Hayes said. Third-party replication would be cost-prohibitive.
The most critical applications in his environment are asynchronously replicated within the ESX cluster, so the target server is usually about 15 minutes behind the primary machine. In the event of an outage, administrators use Citrix to access the target server. The data from the target server is also backed up to Overland disk at the secondary site.
Hayes acknowledges this isn't for everyone. "We don't have very stringent recovery requirements," he admits. "We're not like an online retailer -- bringing another site up in 24 hours is fine." While it's serving his needs for now, he's closely investigating the Overland replication tools as a replacement. The VMware approach is less costly than a full-blown replication tool, but it's not without its own costs. Taking advantage of built-in replication capabilities with the VTL "might be a better way of getting secondary data off site in addition to the most critical."
VMware's parent company EMC acknowledges that stretched clusters can be a viable configuration for disaster recovery, but there are many caveats, according to a blog post by EMC senior director for VMware strategic alliance Chad Sakac.
"In disasters, split-brain can be possible," Sakac wrote. "A bigger deal is the 'smoking hole' disaster. If [you stretch the storage fabric and have the storage at one site], you're hosed (the single array is toast). [Or you could] face a long, painful and operationally complex rebuild."