Tracy

Sometimes I feel like a Caltrans worker late at night. I'm not sure that I agree that "Maintenance really is easy." There have been some architectural changes that had to be performed in-flight that I called many things but easy wasn't one of them. Things are really interesting with so many public APIs changing so frequently. - Tracy

after a hard day on the boat - Tracy

Nice old post on writing bash shell scripts. - Tracy

Better (IMHO) advice: write any non-trivial script in Python (or Perl, if you swing that way). - Tudor Bosman

I think it depends on which layer of the system I'm working on and my perception of how heavily/tightly modeled I want this piece -- which depends on how long I think I will need the piece and how much I will build on top of it. If I need to manipulate the system via command line tools for a quick hypothesis or some ops emergency I rely heavily on bash. Pieces of the system that are more long-lived custom ops functionality tend to be Python (and then sometimes bash scripts rely on python programs), for application logic/production code I like Java (static typing + tool infrastructure) and for data intensive operations C and as little C++ as possible. And Javascript for UI. Perl is the devil. - Tracy

Called the “Plain Preferred Term Sheet,” the document (embedded below) was inspired by a recent debate sparked by entrepreneur Chris Dixon (co-founder of Hunch) and investor Fred Wilson, who have been seeking a way to simplify the complicated provisions that have crept into the average term sheet. - Tracy

OAuth proxy for those that use the command line. - Tracy

"As researchers rush to commercialize printable batteries that pattern organic semiconductors onto paper-thin, flexible substrates, a German team claims to be on-track for a 2010 product launch. " - Tracy

A common concern with the current crop of Solid State Drives is the performance penalty associated with block-rewriting. The flash memory used on today's SSDs is comprised of cells that usually contain 4KB pages that are arranged in blocks of 512KB. When a cell is unused, data can be written to it relatively quickly. But if a cell already contains some data--no matter how little, even if it fills only a single page in the block--the entire block must be re-written. - Tracy

DIY board games. - Tracy

Slides and video for Voldemort, Cassandra, Dynomite, HBase, Hypertable, CouchDB, VPork, MongoDb - Tracy

ouch. - Tracy

The sad tale of sound in Linux. - Tracy

interesting background, though much has since changed - Mike Chelen

It's true, much has changed, but not necessarily for the better. Pulseaudio _promises_ to fix various issues and become the standard Linux sound server ... but at the moment it's still a source of grief for many Ubuntu users (myself included). I've followed the evolution of Linux sound as an end-user since the OSS-only days, waiting, hoping, that it will stabilize - it hasn't happened yet. Linux distros seem to get lots of things right, and so it's puzzling why sound has been so consistently stuffed up. - Andrew Perry

I've solved a lot of bugs while sleeping. Now I just need a REM nap room at the office. - Tracy

on the nature of introspection - Tracy

It tastes so much better than blackberry brambles. - Tracy

Yum ! Mooooo ! Yum ! - Andrew Perry

Oh summer, where art thou!? - ORiGO

Hadoop + Hive is like MapReduce + SQL. - Tracy

"In a number of tests, we observed that our indexing scheme can answer range queries tens of times faster than the well-known indexing schemes." - Tracy

Using bitmap indices, range queries can be answered with bitwise logical operations. Since bitwise logical operations are generally very well supported by computer hardware, uncompressed bitmaps involving relatively smaller number of bitmaps can be efficiently answered. In most scientific applications, the number of bitmaps in a bitmap index is typically large, say more than 1000. This causes the uncompressed bitmaps to be very large. Our compression scheme addresses the size problem by reducing the compressed size to the level that is below a typical B-tree implementation [3]. In addition, because our compression scheme supports fast bitwise logical operations on the compressed bitmaps, the time to answer a one-dimensional range query is a linear function of the number of hits [3]. This is theoretically optimal, since any system to answer a query must at least list all the hits and therefore has the same theoretical time complexity as our approach. This theoretical property is remarkable as only a few most efficient indexing schemes including B+tree and B*tree have it. There are two factors that make FastBit even more remarkable. Since answers to one-dimensional queries produced from bitmap indices can be easily combined to answer multi-dimensional queries, bitmap index scheme that is efficient for one-dimensional range queries is also efficient for multi-dimensional range queries. The same is not true for other optimal indexing schemes. In addition, FastBit is not only theoretically efficient; it is also efficient in tests involving various different application datasets - Tracy

Quick overview of scalable infrastructure components. - Tracy

Time to join the denormalization revolution! - Tracy

"Instead, invest the additional time in crafting a broader suite of web services that access data in ways that are most commonly used, so that the smallest possible payload can be returned." Minimization of payload size is not the only variable. Don't forget latency. Depending on response size, a denormalized response that doesn't require additional calls may be better. The best policy: It depends... - Tracy