Finding China, Crystal, and Tableware With Linux - page 2
A Humble Start
Meredith said it took a solid two years to complete the testing and migration. The first year was just internal testing by himself and Unix Administrator Eric Singer. Selected developers and users were brought in on the second year, to iron out any problems. "The goal was to make the migration from Sun to Linux as easy as possible," he said.
They also wanted commodity machines that were more open, could reduce costs, and permitted manufacturer independent hardware choices. Over the years, Meredith and Singer found that all flavors of proprietary Unix had quirks, so scripts always needed tweaking with an occasional re-writing of code. They wanted to avoid having to make changes for new versions.
Early on they did some beta testing with Veritas clustering and Meredith liked the fact that it ran on both Sun and various combinations of Linux hardware. Sun's clustering solution only ran on Sun hardware. He wasn't excited about locking into one vendor and so chose Veritas (Symantec) to run the cluster.
Oracle provided an evaluation package of Oracle 10g RAC, while HP offered hardware needed for testing. Red Hat offered an evaluation version of RH Advanced Server for a three-month trial.
In 2005, Replacements purchased a set of HP Proliant 585s for a two-node Veritas/Symantec cluster to do server-based applications, reporting, and batch processing. They used four 385s as the database servers for their Oracle 10g RAC cluster. Since HP provided the hardware, they also provided first-tier software support.
During testing, Meredith created lots of spreadsheets to show progress and promote the case for Linux with the CIO and senior management. Documenting the fact that Brio (now Hyperion) SQR (structured query reporting) worked reliably, over a three month period, definitely helped sell the project.
Veritas/Symantec was also chosen because their file system was more mature and worked better than the other solutions. During the 2003 to 2005 time frame, Meredith found that the performance was roughly twice that of Ext2 or Ext3. These two had write performance issues, which in turn caused batch-processing speed problems.