Supercomputer concepts reborn in new data server
- By Will Kilburn
ANTs Software's just-announced Data Server 2.0 is a SQL database management system with a twist. Designed for update-intensive applications like stock trading, telecommunications and package tracking, the product is said to be based on supercomputing principles revived from the 1980s that better match the demands of operational databases with the available performance and speed of modern hardware.
"The problem we're trying to solve is the database locking problem -- you can only have one person updating that classic inventory record at the same time," said Gary Ebersole, ANTs' president, COO and CMO. "We look at this problem and say it's still real, it'll always be real."
SQL is well established, but a lot of things have changed since it was invented. "The core technology was really never designed for high concurrency, and it wasn't designed for wireless applications where you have 240 messages a second from 10,000 users," said Ebersole. "And you can do a lot of tweaking to old technology."
Which is where supercomputers come in. Formed in 1985 as CHoPP Computer Corp., Ebersole's company was dormant for more than a dozen years until it was revived and renamed in 1999 as ANTs, short for Asynchronous Non-preemptive Tasks, the supercomputer-era concept that Ebersole said is the key to getting around the database locking problem.
"In the performance world, there are two issues: latency vs. throughput. In most transactional systems that involve human beings, throughput is more important. You want predictable latency, low latency, but not extremely low latency," Ebersole explained. "But [when] you need to get massive amounts of data through, most database systems choke -- the more workload you put on it in an update-intensive world, the slower the response time becomes and the overall throughput sometimes drops."
ANTs' solution: A message-passing architecture, built on a proprietary micro-threaded engine, within the application space.
"It allows you to be incredibly efficient because you can keep that thread busy all the time," said Ebersole. "And the key to performance is not wasting any CPU cycles doing dispatch services from the operating system, waiting for locks. It allows you to be massively efficient, and efficiency is performance. The shorter the code path, the faster it runs."
Ebersole conceded that Data Server 2.0 isn't, and was never intended to be, a world-beater on the relational database market. Established offerings from big companies are fine, he said, for 90% of developers. They aren't typically dealing with the problems of high concurrency and high workloads. ANTs, he said, is intended for that other 10% of developers.
"Microsoft and Oracle want one tool for all jobs. In their world, they've got the biggest hammer and everything's a nail," Ebersole explained. "We argue that there are problem sets that really require a screw."
For other Programmers Report articles, please go to www.adtmag.com/article.asp?id=6265