lrajlich

I agree with most of what is said, but I think the word "dysfunctional" is incorrect and implies the wrong characterization of these situations. Show me the entrepreneurs who came out of childhood situations where fighting, alcohol and drug abuse are the norm? I would agree that many entrepreneurs come out of childhood situations where their family struggled through chaotic situations (characterized by uncertain future, time spent living on the margin, etc.), but that the key element is that the family can remain functional and successful in these conditions, thus being closer to the exact opposite of what is normally categorized as "dysfunctional". I can point to the many cases of first generation immigrants who go on to be successful entrepreneurs. Their parents, as immigrants, faced exactly this sort of chaotic situation and had to find a way to remain functional.

It's good to see hadoop getting some press attention, it certainly deserves it.

The mention of blueprint got me thinking about CSS frameworks and that there's probably alot of value in using one. What are its strengths? weaknesses?

Love the design. Does anyone have any recommendations for a CSS framework?

Agreed, Eric writes an excellent blog and is based on real world experience.

Coordinated marketing and product launches make sense in the context of a Product Development model but cut into your ability to learn in the context of a Customer Development model. Instead of allowing your initial customers to teach you about your product, you are skipping that step and going straight to teaching the market about your product.

I personally use "iostat -xk 1" which shows the values in kilobytes rather than blocks (easier to read) and updates every second.

Tom's hardware forced all the data to the outer sectors of the hard drive. While this reduces capacity it also cuts down on seek time significantly, which would have a tremendous effect on hard drive performance in a database scenario. It appears they got >100% performance boost in database benchmarks with this technique. See this benchmark result: http://www.tomshardware.com/reviews/short-stroking-hdd,2157-...

I've heard of a number of ways of solving this. I don't like json in blobs since you can't query or index the blob data effectively.

I propose the attribute table pattern, which I've used to decent amount of success. It's not as sexy as ff's, but it is simple, easily cached, and queried. Here's an example schema of attributes you could set against a userId.

CREATE TABLE `Attributes` (

  `userId` int(11) NOT NULL,

  `attributeId` int(11) NOT NULL,

  `charValue` varchar(512) default NULL,

  PRIMARY KEY  (`userId`,`attributeId`),

) ENGINE=InnoDB;

CREATE TABLE `AttributeMap` (

  `attributeId` int(11) NOT NULL auto_increment,

  `attributeName` varchar(64) NOT NULL,

  PRIMARY KEY  (`attributeId`),

  KEY `attributeName` (`attributeName`)

) ENGINE=InnoDB;

These predication markets are garbage. Future expectation of a commodity price can be expressed using futures and this is the generally the way in which a trader (or collectively, the market) expresses a view on the future price of the commodity.

And, combine that with arbitrage pricing of futures and you'll see that the current price of your commodity, in this case, the Dow index, includes these future expectations in its current price.

R is a different language (which is vectorized and partially functional) and has a more comprehensive stats library and better for "statsy" things, whereas matlab is better for more "mathy" things like linear algebra.

R is essentially an open source S-Plus implementation. S-Plus licensing can be an issue, particularly when you are running on a cluster. I worked at a place that used S-Plus and they had a to limit the number of S-Plus slots to run jobs against because of licensing.