High Availability Without The High Cost

Rion Morganstern, director of information technology for PCUpgrades.com, a division of integrator and distributor Spectrum Trading (Monterey, CA), likes to keep in touch with his customers. Morganstern has found that sitting down with clients to discuss their concerns often results in additional sales. That was certainly the case recently when he met with one of his biggest customers, Lawrence Livermore National Labs (LLNL) (Livermore, CA). LLNL is operated by the University of California for the U.S. Department of Energy and performs nuclear and weapons testing.

"This was your typical business scenario," said Morganstern. "LLNL was an existing customer. One day I went to lunch with my salesperson and a couple of people from Lawrence. We were talking about their IT goals and some of the problems they were having. One of the themes that came through very strongly was that they were tired of paying astronomical costs to maintain the monolithic RAID [redundant array of independent disks] systems they were using. Not only were these systems expensive to maintain, but they also had limited redundancy and availability. If anything happened to one of these systems, or even if a system just needed routine maintenance, there were availability problems."

Clusters Provide Needed Availability
While discussing these problems, Morganstern suggested LLNL try a solution incorporating clustered Linux servers. Morganstern believes the clustering technology has come of age and can now be acquired relatively inexpensively, considering the features (availability and failover) it provides.

"Our goal was to create a high-availability server cluster in which all components are redundant and provide failover protection," he said. "LLNL is mission-critical 24/7. There are always scientists and technicians in the labs running tests and operations that can take days to complete. If the server they are working on goes down during that time, they can lose countless man-hours of work."

Morganstern believed he could stabilize the LLNL environment by installing a system that is redundant from the hard drives all the way through to the computing environment. If a failure were to occur, scientists are able to work through the failure and move on to the next task. All of the time invested in a test when the failure occurred would not be lost.

After deciding on the type of solution that was necessary, Morganstern went to the drawing board and came up with a proposal using StorCase (Fountain Valley, CA) InfoStation enclosures with RAID controllers and QLogic Fibre Channel switches. The final component was Dell servers that were loaded with Linux.

Morganstern said he chose the InfoStation because of its price/performance ratio. The InfoStation has nine bays and a backplane design using SCA (single connector attachment) drives. Morganstern has used InfoStations in the past and found them to be a cost-effective way to provide a stable and scalable solution.

Proving The Advantages
The biggest challenge Morganstern faced was proof-of-concept - proving to the customer that the solution could do what was required. "It was a pretty simple problem," he said. "We needed to prove this solution could meet the availability requirements of LLNL and be integrated into their environment. We were able to prove to them that if they were transferring a file and the server died, that file would continue to transfer but it would be transferring to the second server." Morganstern spent about two weeks performing proof-of-concept testing in his own labs. The entire installation was done in a day.

Morganstern believes the low cost of clustering solutions make them a viable option for VARs and integrators that have customers requiring high availability. "Enterprise companies tend to purchase large systems from companies like Sun and LSI Logic," he said. "But if one of those systems goes down, the resources on that system would not be available until that computer was brought back online. Also, these large systems are very expensive, and would normally cost customers $1 million to $1.5 million. Providing high availability and failover protection with clustering, using StorCase RAID systems and Linux servers, now costs less than $150,000."

Morganstern believes two factors converged to bring down the price. "Storage vendors like StorCase have been releasing higher quality products at lower prices," he said. "Those products have features that are comparable to the features on the more expensive products from Sun, LSI Logic, and Hitachi Data Systems. Granted, you do not get full enterprise features, but those features are not necessarily what the customer needs." The second factor is the Linux operating system itself. "With UNIX systems, there is a high cost in the operating system itself, the software to run it, and the licensing agreements, which are usually annual contracts," said Morganstern. "Linux is essentially free. The clustering software is not free, but the cost of it is minimal at around $2,500."

Referrals Generate Sales
Morganstern expects to sell a lot more clustered systems going forward. In fact, LLNL has already purchased a second system for use in another department. That sale resulted from a referral generated by the initial installation. Now Morganstern is trying to get other customers to see the light.

"There is no typical customer for this type of installation," he said. "It is a good fit for just about any company that has mission-critical data and will be crippled by a computer going down for any length of time. A server failure can put a company down for hours or days. With a clustered solution, there is no downtime at all. This is a solution that many of your customers will be looking for."