The New Hope, Penn., company six months ago began a migration to Oracles Application Server 10g on Linux, Database 10g, RAC (Real Application Clusters) 10g and Enterprise Manager 10g. At this point, Talk America has moved its decision support system, the first of a host of homegrown applications that will eventually migrate, to the clustered environment. Next up are the companys OLTP (online transaction processing) applications.
Those early system-migration choices say a lot about the kind of company that goes for high availability in a clustering environment: Talk America is "very information-centric," and the decision support system helps management keep its eyes on the business, said Talk America CIO Timothy Leonard. Hence, failed reporting systems are not a choice.
Leonard said that the company saw grid computing as appealing both from an availability standpoint and because of the ability to have a single set of data that multiple nodes can pull from, again underscoring the importance of reporting.
Talk America is running six nodes on two Hewlett-Packard Co. Superdomes, with EMC Corp. hardware. The information-hungry business is able to run two application servers connected to two nodes with full failover, a consolidation thats offering "big savings" in terms of disk space, Leonard said.
"Were definitely seeing results as far as disk space goes," he said. "Im not replicating as much data around by consolidating. Thats a big savings. Were going to get a lot of disk back, after we fully convert. I feel were going to get a real ROI there."
Before the move, Leonard said, both disk planning and data replication were huge headaches. "Disk planning is always an interesting task around here," he said. "Youre just trying to keep up. Were in line with the business units, and theyre just wanting to know more and more about customers: every touch point, every interaction. Business intelligence is key to our business."
Back in the days of 50 Informix databases, Talk America found that queries would often freeze databases. Database lockups in turn affected OLTP applications, which presented potential data issues. There were mirrored instances, but failover was anything but seamless, Leonard said.
"If there was a long transaction in a [database instance], that instance would be frozen," Leonard said. "If we needed to query, wed have database lockup that would affect OLTP applications. Things like failover werent as seamless as they are with grid. We replicated the database, sure, but the failover wasnt seamless. If we did have an outage, we would have to make the decision to move over to that other instance. And potentially, if replication werent up to speed, wed have a data issue going forth."
Hence, mirrored instances were used more for disaster recovery, as opposed to real-time failover. "We were pretty good in queuing things up, but we did a lot on the application side to prevent data loss," he said. "We didnt have a data loss, but we feel this new architecture will give us not only availability but performance."
Grids seamless replication has solved that problem, and its also eliminated the replication headaches, Leonard said. "We were doing a lot of replication, moving data from the decision support system" with the earlier setup, he said. "This way, we can access production data without affecting production, from a reporting standpoint."
Has the migration been seamless? So far, Talk America has seen "pretty clear sailing," Leonard said, with Oracles support staff getting high marks.
One issue Talk America has run into has been with tweaking homegrown applications to run on the cluster. This has amounted to little more than tuning, however, and has been more of a "knowledge transfer" from Oracles development team to that of Talk America, Leonard said.
All businesses may not be so lucky, however. Charlie Garry, an analyst at Meta Group in Simsbury, Conn., pointed out that some applications flat out will not run on RAC. "If your application is I/O-bound, RACs not going to help you, necessarily," he said. "Theres a bottleneck there, and it has nothing to do with the database."
RAC, in conjunction with data partitioning, could help, Garry said. In fact, people trying to run warehouses with RAC invariably use RAC and data partitioning together. The one barrier: cost. "Youre talking $70,000 per processor to buy partitioning, RAC and enterprise licensing," Garry said.
Still, most people who are doing RAC already own their Oracle licenses, Garry said, and so are not making a huge investment upfront. And, because they consolidate processors, annual maintenance costs can be drastically cut. One Meta client running on RAC went from paying over $500,000 in annual maintenance fees down to $100,000, all thanks to running on fewer processors, he said.
What Garry sees as most exciting about RAC, however, has to do with the fact that customers are increasingly citing not just the lure of high availability that Talk America is happy about—theyre also talking about scalability. Oracle has long promoted the benefits of scaling on low-cost commodity servers as being a big plus of its grid computing architecture, but few if any customers could be found that were running on more than a few nodes.
Thats changing, Garry said. "Two or three years ago, I saw people running RAC on two-node clusters, clearly for high availability," he said. "Nowadays, Im running into people—not people Oracle sent me to, but clients—running four, eight, 16 nodes. Its not too uncommon."
For Talk America, however, scalability is not an issue, considering the heavy-duty equipment on which its business runs. "Were not using low-end equipment," Leonard said. "These are hefty machines with heavy memory. Another key to running databases is heavy I/O channels. We run a lot of fiber I/O channels to EMC equipment. … Performance is not an issue at this point or in the foreseeable future."