b0b0b0b

I think the real wtf was testing in production. At least their lab now has an environment to do their qa on.

This reminds me of Eclipse's support for native hooks to detect file updates (in case something was changed from outside the eclipse session). Watchman would then be like a deconstructed ide.

Do the bigtable performance characteristics look kind of like cache line ping ponging? My intuition for scenario 3 outperforming scenario 2 (100% remote vs 50% local + 50% remote) is that there are more mutations of data and therefore more interconnect traffic is required to maintain coherency across sockets.

Try this with perf stat on cpu 0, on cpu non-0, and without taskset at all.

If you have 2-threaded code, it's also fun to try taskset onto one core with hyperthreading.

Sounds tantalizing, but in reading it is there a risk of patent infringement liability?

There's some good stuff here too: http://www.cs.sunysb.edu/~skiena/392/

I wonder if Netflix is considering standing up their own data centers, given how much their business relies on Amazon.

The title was groan inducing, but the article has introduced me to rotation forests. Thanks!

Thanks for this interesting writeup; it's definitely thought provoking.

Because you can't index that bloom column, it seem's you'd always be doing full table scans.

In fact it doesn't appear any indexes were used throughout this whole exercise, is that right?

Has he told the story behind the dot? Better than this story IMHO.

If I'm dealing with [compressed] files on disk or across a network, what's the practical overhead of pv? A few extra percent CPU, or some material decrease in through put?

iono; I'm happy to see Parker up there.

9500 servers cost only $5M per year? Sounds like putting a positive spin on evicting outdated, idle boxes. Did I tell you I ate my veggies last night?

Do you have a cluster of application servers? What transaction isolation level do you use? Will User A mind that his update is no longer a single transaction and that it may be commingled with txn's of Users B&C?

For me, to get it to run, I had to create a "configuration" in Models called "DivvyIsomap".

But I got it running and it looks cool. Thanks!

These issues are also why sql hints exist.

Streaming Netflix still works from MAC/PC, but personalized recommendations are absent from the front page.

So I think this is a win in terms of decoupled services and degraded mode functionality.

I expect they probably pushed out some code too hastily due to the recent licensing hoo-hah with the media companies (and can't just roll back because of said hoo-hah).

I usually get in to the office a bit after 8. I try to fix one bug / get one thing Done before 9. After that, everyone else gets in the office and everything descends into chaos and meetings.

I'm interested in learning more about how customers are modelled. Transaction history, # support cases, usage patterns; these will be more informative than acquisition channel. But then there's the matter of getting data into and out of Custora. Looks really cool, guys.

it seems that his algorithm is faster because it exploits the model of computation (memory aligned accesses and multi-byte operations). He gets up to a constant factor more comparisons for free.

These guys got some press several years ago: http://www.wired.com/culture/lifestyle/multimedia/2006/05/fi...

I was amused hearing of fights where they beat each other only with rolled up newspapers. It beats watching tv and eating pork rinds, I guess.

I like to imagine that the bing algorithms are so smart, that it realizes the preciousness of the signal in clicktrack-logged visits to websites with a referrer "http://www.google.com/search?q=%s. This single feature is given so much weight that bing unintentionally gives the appearance of wholesale google duplication.

I wonder if anyone at woot had the groupon idea (before groupon came out) but was quashed because it just wasn't woot-like. It seems like woot lost a couple orders of magnitude of scale by (apparently) making themselves the fulfillment bottleneck.

When modelling hypothetical trades do you account for slippage and transaction costs?

They managed to test and train on different periods of time.

    February 28, 2008 to November 28, 2008 is
    chosen as the longest possible training period while
    Dec 1 to Dec 19, 2008 was chosen as the test period

That said, their split exhibits enormous bias:

    Dec 1 to Dec 19, 2008 was chosen as the test period
    because it was characterized by stabilization of DJIA
    values after considerable volatility in previous months
    and the absence of any unusual or significant socio
    cultural events

I don't understand why they would think this is okay.

Google seems to attack this problem from a variety of expensive angles [1], [2]. But in the end, the simplest thing sounds best to me [3].

[1] http://glassbox-design.com/2009/google-goes-aquatic-photos-i...

[2] http://www.engadget.com/2010/06/17/another-reason-to-want-a-...

[3] http://www.reddit.com/r/pics/comments/6ceia/google_naps_phot...

collaborative ad filtering

he has a multimonitor setup

git-flow reminds me so much of maven.