Oracle Gets Zippy with Financial Services - Page 2
Thats had a domino effect in this business, Iati said. Its increased the activity and amount of information that has to be moved, based on activity that could happen at penny increments as opposed to 16th of a dollar increments. To give you an idea of what kind of volume increase this translates to, Iati said that, over a period of six years, peak message rates in the industry have gone from 4,800 messages per second to 55,000 mps.Obviously, given all these needs, database companies are eager to step in. Figuring out how to get data in and out of the system really, really fast isnt easy, however, Olofson said, given the constraints of a traditional database.One technique used is to create a caching environment that can be used interactively with database clients, instead of being forced to go back to the full database itself. Weve also seen approaches that involve clever uses of buffering to achieve these things, usually coupled with a partitioning scheme so you can manage data in buffers, thus ensuring that a given transactional query will be in memory. What Makes TimesTen So Fast? TimesTens approach is different from these. It uses main-memory database technologyas in, theres an actual database resident in main memory, not just a buffer or structured cache. Even TimesTens Cache product is still a database; its just that people use it as a cache, given that it has facilities that link it back to the database. The fact that TimesTen is a full database, as opposed to an approach that utilizes cache, makes it a better solution, Olofson said. Specifically, it spools out its own log, which assists in data recovery. "When people talk about main-memory databases, people have the concept that its just memory, so if the system goes down, everythings lost," he said. "Thats not true [with TimesTen]; it spools transactions out to logs. If it has a failure, it can recover from the log, [which] saves snapshots at intervals." The distinguishing characteristic is that the TimesTen database doesnt have to go to the disk. A disk-based database is optimized around the rapidity with which you can randomly store and retrieve data to disk, with memory and instruction structure in the database kernel all organized around those issues. Main-memory databases couldnt care less about that issue. With database in memory, theres no shuffling of stuff. TimesTen claims to be able to move data 10 times faster than standard disk-based databases, even when the disk-based database has all its data in memory. Next Page: The acquisition will change the competitive landscape.