Monthly Archives: September 2007

T-Minus a few days until launch

Not so many posts lately but it’s not due entirely to laziness. I’ve been busting my butt getting my site ready for public consumption. I built a server by hand which was fun, it was sort of like putting a Ferrari engine in a Peugeot. I bought a few extra fans so it should last long enough for a launch and a few looks from investors. We have a venture capitalist stopping by in a week or so to have a look, fingers crossed.

I had a long, interesting conversation with a postdoc “wet” neuroscientist yesterday. Interesting guy, he has a similar vision but wants to apply it to the mess that is scientific publishing. Wet means he actually pokes and prods mice brains. We’re going to meet up with another guy at Caltech in a week or so to discuss collaboration potential.

I left work today exhausted but really happy. I really like my co-workers. I’m more productive than I’ve ever been, and I’ve been afforded the opportunity to build my idea.

The Last Bottleneck

Geeky braindump ahead. Thoughts on building a high performance server for use with the site I’m launching on a budget.

I have this weird obsession with solid state drives because they are the final bottleneck in performance that can prevent someone from cheaply creating a scalable server to host an idea accessible by millions of people. Modern CPUs and dual channel DDR RAM are sufficient to host most websites, the problem is the expense of large high performance RAID arrays required for database driven websites. My hope is that some day a kid can write the next big app in his house and host it on his home PC with the only additional expense being a good internet connection. Heck you may even be able to host a WordPress blog and get some decent traffic without it crashing.

Prices are dropping fast but RAM based drives like the 16GB (sufficient for most web apps) Hyperdrive4 pictured above are not yet able to take advantage of SATA2. So they’re stuck at 140MB/s unless you build a RAID array which is prohibitively expensive. The next generation Hyperdrive will support 3.0Gb/s (roughly 280Mbytes/s factoring in overhead). That’s when things get really interesting because a single Hyperdrive would blow most super expensive RAID arrays out of the water at 3.0Gb/s even when not using RAID.

So, in theory, if you used Amazon’s S3 image hosting, you could run a server with the following specs and avoid most bottlenecks assuming your web app is well designed. You can build this system on NewEgg.com right now for under $700 without a hard drive..

Intel Core 2 Duo at 2.4Ghz
4GB DDR2 RAM
Intel 945 based Mobo / Barebones Kit

Right now a 16GB Hyperdrive will set you back $2400 + $2,000 for the RAM = $4,400. So a total of $5,100 for the whole “server”. That will get you a pretty decent Dell server but it would be bottlenecked by IO, even with a good RAID array. Granted it won’t be as reliable as a Dell Poweredge but then your server would have no moving parts so throw in a cheap UPS and some ventilation and the thing should be pretty stable.

My plan is to build the sub $700 server and add a decent PCI-Express RAID card and run a pair of WD Raptors in RAID 0. Highly unreliable, like my wacky custom turbo WRX, but it should also be sufficiently fast to handle small spikes in traffic. The cost of the Raptor RAID setup is roughly $700 or about $3700 less that the Hyperdrive setup. So I’m planning on building a $1400 server. If my site takes off and I can afford to upgrade the server my first move will be to plug in a Hyperdrive4 and use the Raptors as RAID 1 backups.

The benefit of RAM drives over your run of the mill flash based drive is write performance. Flash drives, while solid state, have notoriously terrible write performance. RAM drives have no such problems. Also, due to the lack of moving parts, SSDs have ridiculously low latencies, which is crucial when you’re doing random reads and writes. In fact, now that I think about it, the fact that the Hyperdrive isn’t SATAII may not be much of a problem because latencies are already extremely low.

And in case anybody is wondering, this blog currently lives on a Dell PowerEdge with a Quad core processor, 4GB of RAM and a RAID 10 Array which I benchmarked at over 100MB/s sustained. Which is why it feels sorta snappy.

Hot in the City


I managed to get moved to LA and wow is it hot today. 107° F | 74° F according to Wunderground.com. It’s 94.3 degrees at 9:18AM.

Luckily my new place has AC and a friendly dog to keep me company. We’re hooking up a T1 for hosting the site so the site should be hitting the interwebs in about a month or so. I’m really really looking forward to launch day, getting feedback from users, getting hacked and learning why, seeing the first discussions emerge. I don’t have very many ideas that I can actually test out in the real world. My blog has had sporadic success but there is a big difference between writing about an idea and actually building the tool that allows an idea to happen.

Wanted to summarize my current econo-thinking before I forget. Some interesting comments from Rich in the Calculated Risk comments:

The problem with Leamer, the Economist analysis and almost every other think-tank studying this problem is that they think in linear terms, starting with the housing/subprime problem and working from there sequentially to evaluate how much damage it will do or whether it will be contained.

But it’s possible that other structural deficiencies in the U.S. economy could trigger weakness or even a recession, housing apart. Nobody seems to have a multi-factor model or mindset that evaluates how a variety of short-term and long-term economic deficiencies may inter-relate.

Individual posters on this board seem to have that mindset, and the collective board as a whole does a good job arraying and evaluating multi-factor economic inputs. But professional economists are too shallow, timid, and just aren’t doing their jobs very well. That’s why I put more faith in what this board collectively thinks than any economist like Leamer.

The internet is a weird thing, I can ask questions to Hedge Fund managers, retired CEOs, media critics, economics professors, and various other experts and they actually answer. Mindboggling. Anyway, my brain has assembled the following web of ideas in an attempt to understand the world.
Austrian School Economics – Correct but doesn’t account for technological change. That’s a bad idea in a world suddenly subject to the Internet.
Joeseph Schumpeter – Understood business cycles, coined the term “Creative Destruction”, maybe predicted the impact of the internet prior to its existence. Predicted that capitalism would eventually piss off enough people in a democracy that its collapse was inevitable.
Elliot Wave Theory – Too dependent on decades of static behavior. I tend to think business cycles and change reset the “wave” if they’re big enough. Though who knows.
Accelerating Change – There aren’t a hell of a lot of economists that can talk about this topic, which is too bad because as Rich mentioned it’s crucial to economic forecasting.
Emergence – The main reason we still have Socialists is due to a lack of understanding of emergence. It seems like most people don’t understand why capitalism works. Free markets are seen as playground for greed reserved for the wealthy.

I have this feeling that the competitive and collaborative market for ideas allowed by the internet will create a better understanding of economics and business cycles which will lead to better predictions, and better predictions will lead to the death of animal-spirit induced recessions. So instead of wild swings in GDP growth that the Fed is unsuccessfully trying to battle, people will simply be aware that the Fed Funds Rate is too low and that the current trendy bubble will inevitably pop. The knowledge would hopefully lead to slow but steady growth. Instead we have economists proclaiming that there is know way to know if a bubble exists until after it pops, which is of course ridiculous unless you have a vested interest in ignoring fundamentals.

Interesting read:
http://blog.inman.com/LeamerHousingandBusinessCycle.pdf

The dog photo is by Rion on Flickr