FRAMINGHAM (02/04/2000) - The place: Loveland, Colo. The date: Dec. 28, 1999. A freight train believed to be carrying nearly 20,000 gallons of hazardous material has just derailed, putting hundreds of citizens in the vicinity in danger.
Fire Chief Dick Minor has little time to act. Using a computerized map of the city, Minor and his associates quickly outline an area in the shape of a polygon around the center of the crash site.
Within seconds, a computer in the fire department's main transmission center automatically starts phoning hundreds of residents living within that polygon with a prerecorded message asking them to tune in to the city's AM radio station for evacuation and other information.
Fortunately for the 52,000 residents of Loveland, the Dec. 28 "accident" was just a simulation. But had it been real, Minor would have been able to call the affected residents thanks to a recently installed Early Warning Evacuation (EWE) system.
Making EWE particularly interesting is that while reliability is absolutely key to its success, it's entirely based on Windows NT Servers, an architecture many information technology managers say is less than bullet-proof.
SCC Communications Corp., one of the largest providers of 911 services in the country, usually runs all of its core services on highly fault-tolerant servers from Tandem Computers Inc. in Cupertino, Calif., said Mark Knox, a SCC system architect.
Boulder, Colo.-based SCC is also the developer and provider of the EWE service.
The decision to use NT for EWE was dictated by the fact that a crucial third-party geographic information system application needed for the service to work was available only on NT.
"We were very uncomfortable with the idea of running it on NT. We would have much rather had the service running on Unix boxes," Knox said.
Instead, the company has been pounding away trying to make its Windows NT environment as rock-solid as its Tandem servers have proved to be over the years.
The EWE is a reverse 911 service that lets public-safety officials quickly launch an automatic telephone alert to targeted segments of the population.
Ensuring much of the reliability and service availability is hardware and software technology called Endurance Array from Marathon Technologies Corp. in Boxboro, Mass.
The technology ties four single-processor Intel Corp. PC servers into a single, larger logical server.
The servers are broken into pairs, or "tuples," as Marathon calls them. Each pair consists of a "compute" server that houses the operating system and all applications and an I/O server for all I/O operations, said Craig Jon Anderson, a director at Marathon.
Patented synchronization software keeps the application running in absolute lockstep on each tuple. If one tuple crashes, the application keeps running on the other.
"There are always redundant elements executing all operations at all times," Anderson said. "Think of it as the tires on an 18-wheeler . . . even if one tire blows out, the truck keeps running."
With Marathon's technology, the same application runs on two servers at the same time, he said.
But Marathon's Endurance Array can only be used to tie together single-processor Intel servers, which may prompt SCC to consider other options as its scalability requirements grow, said Rusty Robinson, a Unix/NT administrator at SCC.
SCC's next release of EWE, for instance, will be Web-enabled, and users will be able to access it via the Internet, which will require larger servers.
"We are looking forward to the next release of Endurance in hopes it will support [symmetrical multiprocessing] for our SQL Server 7.0 database," Knox said. If not, SCC will probably look at NT clustering technology in Windows 2000.