Computer scientists at the University of California, San Diego and its San Diego Supercomputer Center (SDSC), Eureka-based Silicon Defense, the University of California, Berkeley, and the nonprofit International Computer Science Institute in Berkeley, found that the Sapphire worm doubled its numbers every 8.5 seconds during the explosive first minute of its attack.

Within 10 minutes of debuting at 5:30 a.m. UTC, Jan. 25 (9:30 p.m. PST, Jan. 24) the worm was observed to have infected more than 75,000 vulnerable hosts. (See an animated image illustrating the rate of the attack.) Thousands of other hosts may also have been infected worldwide. The infected hosts spewed billions of copies of the worm into cyberspace, significantly slowing Internet traffic, and interfering with many business services that rely on the Internet.

"The Sapphire/Slammer worm represents a major new threat in computer worm technology, demonstrating that lightning-fast computer worms are not just a theoretical threat, but a reality," said Stuart Staniford, president and founder of Silicon Defense. "Although this particular computer worm did not carry a malicious payload, it did a lot of harm by spreading so aggressively and blocking networks."

The Sapphire worm's software instructions, at 376 bytes, are about the length of the text in this paragraph, or only one-tenth the size of the Code Red worm, which spread through the Internet in July 2001. Sapphire's tiny size enabled it to reproduced rapidly and also fit into a type of network "packet" that was sent one-way to potential victims, an aggressive approach designed to infect all vulnerable machines rapidly and saturate the Internet's bandwidth, the experts said. In comparison, the Code Red worm spread much more slowly not only because it took longer to replicate, but also because infected machines sent a different type of message to potential victims that required them to wait for responses before subsequently attacking other vulnerable machines.

The Code Red worm ended up infecting 359,000 hosts, in contrast to the approximately 75,000 machines that Sapphire hit. However, Code Red took about 12 hours to do most of its dirty work, a snail's pace compared with the speedy Sapphire. The Code Red worm sent six copies of itself from each infected machine every second, in effect "scanning" the Internet randomly for vulnerable machines. In contrast, the speed with which the diminutive Sapphire worm copied itself and scanned the Internet for additional vulnerable hosts was limited only by the capacity of individual network connections.

"For example, the Sapphire worm infecting a computer with a one-megabit-per-second connection is capable of sending out 300 copies of itself each second," said Staniford. A single computer with a 100-megabit-per-second connection, found at many universities and large corporations, would allow the worm to scan 30,000 machines per second.

"The novel feature of this worm, compared to all the other worms we've studied, is its incredible speed: it flooded the Internet with copies of itself so aggressively that it basically clogged the available bandwidth and interfered with its own growth," said David Moore, an Internet researcher at SDSC's Cooperative Association for Internet Data Analysis (CAIDA) and a PhD candidate at UCSD under the direction of Stefan Savage, an assistant professor in the Department of Computer Science and Engineering. "Although our colleagues at Silicon Defense and UC Berkeley had predicted the possibility of such high-speed worms on theoretical grounds, Sapphire is the first such incredibly fast worm to be released by computer hackers into the wild," said Moore.

Sapphire exploited a known vulnerability in Microsoft SQL servers used for database management, and MSDE 2000, a mini version of SQL for desktop use. Although Microsoft had made a patch available, many machines did not have the patch installed when Sapphire struck. Fortunately, even the successfully attacked machines were only temporarily out of service.

"Sapphire's greatest harm was caused by collateral damage-a denial of legitimate service by taking database servers out of operation and overloading networks," said Colleen Shannon, a CAIDA researcher. "At Sapphire's peak, it was scanning 55 million hosts per second, causing a computer version of freeway gridlock when all the available lanes are bumper-to-bumper." Many operators of infected computers shut down their machines, disconnected them from the Internet, installed the Microsoft patch, and turned them back on with few, if any, ill effects.

The team in California investigating the attack relied on data gathered by an array of Internet "telescopes" strategically placed at network junctions around the globe. These devices sampled billions of information-containing "packets" analogous to the way telescopes gather photons.

With the Internet telescopes, the team found that nearly 43 percent of the machines that became infected are located in the United States, almost 12 percent are in South Korea, and more than 6 percent are in China.

Despite the worm's success in wreaking temporary havoc, the technical report analyzing Sapphire states that the worm's designers made several "mistakes" that significantly reduced the worm's distribution capability.

For example, the worm combined high-speed replication with a commonly used random number generator to send messages to every vulnerable server connected to the Internet. This so-called scanning behavior is much like a burglar randomly rattling doorknobs, looking for one that isn't locked. However, the authors made several mistakes in adapting the random number generator. Had there not been enough correct instructions to compensate for the mistakes, the errors would have prevented Sapphire from reaching large portions of the Internet.