Crashes Again On Monday

For the second consecutive workday, suffered a major crash on Monday (June 9), with the increasingly unlikely scenarios explaining why the historically robust site is failing.

The cause of the crash, which apparently took the weekend off after bringing down Amazon completely on Friday for almost three hours before seriously (but less severely) slowing down Amazon for several hours on Monday, was unclear.

For itself, the question of what caused the crash was one of those good news/bad news situations. Good news: Amazon knows what caused the crash. Bad news: They won't say. (This appears to be the closest Amazon plans on getting to issuing a statement explaining the outage.)

"We're not commenting as to the cause of Friday's or Monday's site issues. As a policy, we don't discuss such matters," said Patty Smith,'s director of corporate communications. "We are aware of the cause. We're just not disclosing the issue."

Asked about some of the reports of various scenarios for the outage, Smith said they weren't going to go there. "I've seen several comments from Keynote and other third parties speculating as to the cause, but I'm afraid I'm not going to confirm or deny any speculation," Smith said. "Suffice to say, that on those rare occasions when our site experiences problems, we work to resolve the issue as quickly as possible."

I quoted that last line only because it's such a classic political line, where the candidate throws in a mom-and-apple pie line as though someone has challenged him on it, even though he was the one who brought it up. "I can't really comment about those undercover FBI tapes of me taking bribes, but I will stress that no one loves my constituents as much as I do. And I vigorously refute anyone who says that the 13th Congressional District is made up of anyone other than kind-hearted, wonderful people."

In short, no one has suggested that Amazon was being slow to fix this issue. The concern is with not saying what caused it. That's just the kind of confidence-building message Amazon wants to send out to its suppliers and to signal to its customers. Why spoil that wonderful "this could happen again tomorrow" surprise that is at the heart of E-Commerce enjoyment?

Web site performance tracking firm Keynote said that Friday's crash was most likely due to an overly sophisticated site operation at Amazon. In a sea of unlikely scenarios, that is probably the least unlikely. Amazon has historically deployed the most advanced capabilities of any major E-Commerce site, which could set in motion a domino effect, with a small error anywhere in one app potentially cascading into a major crash.

The only problem with that theory is that Amazon's systems have had such a level of sophistication for years, maintaining an almost spotless record of high availability. Why would it suddenly crack now? The most likely answer is that they have recently added some new piece of software that brought with it the instability.

Another possible explanation is excessive load—some new game programs were released on Friday, and Monday saw Apple updating its popular iPhone—but that's the kind of thing that Amazon has handled quite well historically.

Some reports suggested some kind of attack, especially a Denial of Service assault. But Shawn White, Keynote's director of external operations, said the data he's been monitoring simply doesn't fit the DOS pattern.

"When that (DOS) kind of situation happens, the site slowly gets slower and slower. That's not what happened here," White said. "DOS doesn't look likely at all."

But that still theoretically leaves open the possibility of some other form of attack, possibly some sort of planted malware. There's no hard data available thus far, however, to support that theory. The only information that White saw that fueled the malware rumors were reported slowdowns at eBay and the Internet Movie Database, but those could easily have been unrelated.

White is sticking—for now—with his complexity theory. "As a site becomes more and more complex, each moving part becomes more and more dependent upon every other part," he said. "If one piece has a mistake in it, a typo in it," that could potentially trigger a collapse.

Monday's Amazon crash differed from Friday's in a few respects. First, it was a major slowdown on Monday, as opposed to the total crash on Friday.

The Monday incident started at 1:03 PM (EDT) when, according to Keynote, Amazon's homepage plunged to 30 percent availability before recovering completely 20 minutes later. At 1:56 PM, its availability again dropped, this time to 68 percent, before recovering at 2:09 PM. It dropped again to 68 percent from 3:43 PM until 4:01 PM.

Another key difference: Friday's crash impacted only the main site, with no apparent impact on any of its non-U.S. sites or its cloud computing Amazon Web Services site. Monday's crash also hit Amazon's United Kingdom site, Keynote's White said.

"The problems experienced on this site lasted longer and were, in some cases, more dramatic. Beginning at 1:06 PM (EDT), the same 'Service Unavailable' error was being seen by European online shoppers," White said. "Availability dropped to 30 percent and then slowly, over the next couple hours, returned to normal. By 3:02 PM (EDT), the UK site returned to normal."

The UK site also acted a little different as shoppers dug more deeply into the site.

"During each outage period, visitors who were able to make it past the homepage and browse for items would have experienced much slower performance and download times. Both sites regularly download completely in less than seven seconds and in many cases, faster than four seconds," White said. "Download times during the periods mentioned slowed by as much as 200 percent. In some cases, Web browsers would have completely timed out, causing visitors to have to reload their browsers, unsure if the site would return."