Amazon: Insight into Search, Engineering, and Cloud Computing

April 28, 2011

In order to locate data, one must be able to search for it. If search does not work, data are lost. Seems obvious but one of the consequences of the Amazon cloud outage was that I had to think about the online big box store again. Amazon is, to me, a convenient way to get books and buy a gift or a replacement BlackBerry battery. Even when the A9 service was a priority, Amazon’s ability to make information findable was hit and miss.

Even today, I have a tough time thinking of Amazon as giant, reliable, low cost information utility. I have difficulty finding lists of books “about” a subject. Sometimes I stumble upon this user created content; other times, I have no idea how to find this useful information. When I want a book, I don’t know how to NOT out books that are available from those that will be published in the future. I cannot find information about the credits I “earn” when I buy Kindle books or products using my Amazon credit card. The snail mail coupons I used to get have disappeared, and I don’t have a clue about “finding” this information.

Several years ago, we did a close look at how Amazon handled glitches. The information was not that different from other companies we had examined. However, one approach was interesting. When an outage took place, a small team was assembled to figure out what happened and to fix it. This approach has its upside such as speed and fluid problem solving. The downside, in my opinion, was that solutions could be ad hoc. In my view, the next time a problem cropped up, the Amazon approach I probed three years ago meant that the next problem solving team had to figure out what the previous team did. No big deal until the problem of figuring out everything consumed lots of time.

We are not using Amazon Web services. Call me old fashioned but I prefer to have data storied on local devices with appropriate backups on media in an off site location.

For another, unrelated project we ran a series of tests in 2010 on the take up of the phrase “cloud computing.” What we learned was that the actual traffic generated by the phrase “cloud computing” was far less than our client anticipated.

After a six month text, we concluded:

  • There was a large amount of information about cloud computing from a bewildering range of vendors big and small
  • The interest in cloud computing was less than in some other words and bound phrases we tested
  • The information about cloud computing was a cloud of semantic fuzziness; that is, it was difficult to pin down specifics within the documents written about cloud computing.

What happens when you combine a retail store with a cloud computing service? You get an anchor point. Amazon becomes associated with certain words and phrases, but these may not have much meaning. Examples range from acronyms from S3 to EC2.

What happens when a company which has associated itself with this difficult to define subject has an outage? The problems of Amazon immediately diffuse across other products and services available in the cloud.

You can see an example of this semantic drift in “Amazon: Some Data Won’t Be Recovered after Cloud Outage.” The article points out that the Amazon “outage” has resulted in data that “won’t be recovered.” The problem is no one that Amazon and its customers must resolve.

Amazon’s close association with cloud computing has made the Amazon incident the defining case for the risks of cloud computing. Even worse, unrecoverable data cannot be found. Search and retrieval does little good if the data no longer exist. Services which depend on their customers locating information are effectively stranded. Those affected include “Quora, Sencha, Reddit, and FourSquare.”

So what?

This problem at Amazon provides some insight into the firm’s engineering approach. In a larger arena, the close association of Amazon with cloud computing has had a somewhat negative impact on the concept of cloud computing. To sum up:

  • You can’t find information if it is not  “there”
  • Amazon’s engineering methods are interesting and may give some companies some additional analysis to perform
  • The impact of the outage has created some pushback for other cloud computing vendors.

Will this be a defining moment for Amazon? Probably not, but it is an interesting moment. Non-recoverable is a disturbing notion to those who have to find a fact, entity, or a concept. Amazon has figured out some aspects of eCommerce. Other areas warrant additional investment which may be why Amazon’s costs are skyrocketing.

Stephen E Arnold, April 28, 2011

Freebie

Comments

Comments are closed.

  • Archives

  • Recent Posts

  • Meta