Tom Vanderbilt, author of the excellent Traffic, has a very good piece in the latest NYTimes Magazine on data centers.
The specter of infinitesimal delay is why, when the Philadelphia Stock Exchange, the nation’s oldest, upgraded its trading platform in 2006, it decided to locate the bulk of its trading engines 80 miles – and three milliseconds – from Philadelphia, and into NJ2, where, as Thomas notes, the time to communicate between servers is down to a millionth of a second. (Latency concerns are not limited to Wall Street; it is estimated that a 100-millisecond delay reduces Amazon’s sales by 1 percent.)
…It seemed heretical to think of Karl Marx. But looking at the roomful of computers running automated trading models that themselves scan custom-formatted machine-readable financial news stories to help make decisions, you didn’t have to be a Marxist to appreciate his observation that industry will strive to “produce machines by means of machines” – as well as his prediction that the “more developed the capital,” the more it would seek the “annihilation of space by time.”
I like the quote but doubt that Marx is the best guide to this new world. try Charlie Stross instead.
it is estimated that a 100-millisecond delay reduces Amazon’s sales by 1 percent.
Someone is going to have to explain this one to me.
May I say a few words about Accelerando?
come to think of it, this might also be an interesting pricing option for cloud operators. a cloud operator such as Amazon could offer you a price for a certain level of latency with other entities on its cloud and a different price for a lower level of latency. At that point Amazon does not have to coordinate with anyone, but just need to partition its could by latency service agreements and place each partition wherever convenient.
Alex needs to post more. And share his “Assorted links”.
just as an experiment, try delaying everything you say or do by 100ms today, and at the end of the day pass out a questionnaire about your intelligence to all people you have interacted with.
Allan, you do not generally want to locate data centers in hot areas, in fact, the further up north the better – you safe *a lot* on cooling.
You can do without water (unless you plan on stream cooling which the environmentalists wont like, anyway)
@allan
i am sure the navajo would have something interesting to say about the annihilation of space by time
Someone,
Heat can be ameliorated. For example, instead of building up, build down and use the earth as a natural coolant. Further, evaporative coolers (if you can get enough water) are great for cooling things down while using little electricity in dry areas.
But if you want cool, then just choose rural ND, SD, Montana, Wyoming…
Babar wins the thread.
This, in a nutshell, is why humans will never colonize outer space, beyond at most low earth orbit.
Latency issues will only loom larger as networked communication becomes ever more pervasive. Some future analog of Twitter might truly be real-time, with the human race soaking in a permanent ambient warm bath of “tweets” and reacting instantly like pulsating ripples in a school of fish changing direction in unison at subsecond intervals.
Who says anybody has to be conscious of the effect of latency?
There will be no singularity. Speeding up your ability to process garbage does not result in a singularity …. it results in more garbage. And their is nothing more stupid that automated trading models.
The real advantage of computers is not artificial intelligence…we don’t need that for the most part. We have more than enough natural intelligence. We need to automate all the various brainless activities that humans currently do. It seems stupid to try to make computers good at something humans are all ready capable of and not focus on things which computers are incredibly good at and humans are bad at (like calculating taxes).
why increases in latency cost Amazon money:
A company like Amazon has all the tools to process a sale spread across dozens of servers. One probably manages credit card transactions, one records search behavior, one sends a message to shipping, etc, etc, etc. Imagine that a single purchase requires 100 messages between servers. That would be an extra 10 seconds of server time. Multiply those 10 seconds by (at least) thousands of sales a day. This extra time runs mostly concurrently so a user won’t see it, but it would require more servers to process the sales.
(all number pulled from thin air for illustrative purposes)
Comments on this entry are closed.