P16: Practical Progress

Supercomputing for the masses: Nvidia CUDA

This is a post I should have written two weeks ago, but I've been lazy. (Or busy...)

I went to a seminar by Nvidia about CUDA, which is a framework for writing apps to run directly on an Nvidia GPU. Not just graphics apps, either. Nvidia GPUs are massively parallel -- up to 128 hardware threads -- and you can write arbitrary C code to run on it.

The seminar was part of a financial technology conference, so they had a third-party app provider which naturally was building solutions on CUDA. They said they were seeing a 50x speedup (one Nvidia GPU versus one Intel CPU core) when running their compute-heavy applications.

I came home and immediately installed their CUDA SDK for Mac OS X on my MacBook Pro. To my surprise, it ran with minimal hassle (*), and I was promptly playing with particle field and liquid simulations. But my Mac doesn't have enough VRAM to run some of the more intensive demos. (MonteCarlo, BlackScholes)

(*) The only real tweak necessary was: "export DYLD_LIBRARY_PATH=/usr/local/cuda/lib"

Takeaways:

• It's a powerful parallel-programming model, that generates code for the GPU or the host CPU
• You can start developing with commodity Nvidia GPUs, even in laptops.
• You can scale up to higher-end solutions - up to 4 128-thread GPUs per "host" system

One thing CUDA doesn't provide is a way to manage and process massive datasets. The cards have somewhat limited memory, and you have to write an app that runs on the host and feeds the card with data. For search applications like topic clustering -- which I'd like to use this for -- CUDA alone doesn't provide an answer.

Perhaps it would make sense eventually to use Hadoop plus CUDA -- write your map/reduce tasks in CUDA, and rely on Hadoop to distribute data around a cluster of Nvidia-accelerated boxes?

Two spiffy related projects:

• pycuda Python wrapper for Nvidia CUDA
• Mac OS X 10.6 "Snow Leopard" is slated to ship with OpenCL, roughly a version of CUDA that works on multiple GPU architectures

Special thanks to Rich Hecker for posting a heads-up about the CUDA seminar to the NY-newtech list.

Simple SEO Tips

This week, someone asked on the NY Tech Meetup mailing list: "What (are) some of the basic SEO optimizations and techniques you guys have used?"

Craig Wood posted a wonderful, succinct response. I'm reposting it below for future reference.

The only thing I would add: keywords in URLs and hostnames matter. Google ranks keyword-in-URL matches (and title) higher than just keyword-in-document. The closer this occurs to the start of the URL, the better. Hostname match is best of all (e.g. "newyork.craigslist.org")

Or, this was true the last time I saw any serious SEO analysis research, which is ~2 years ago. Maybe things have changed since then. :)

---

1) Titles - The biggest Bang you will ever have -- Never Duplicate your <title> tag

• Happy Valentines Day 2008 from My Site
• My Site - Information About Us

2) Common sense linking -- make all your URLs descriptive

• /news/2008/02/14/happy-valentines-day.html
• /information/about-us.html

3) Alt tags on Images

You do these three things well and you are 95% done with what you can do to get improve your ranking with the HTML of your site. There are all kinds of things you can do to tweak your content to be more towards culture references or 'hot searches'. And of course have a robust (no spam) external linking campaign.

-Craig

Slides on cluster computing and MapReduce

As found on Geeking with Greg:

Google has published slides and video from a course taught to interns during the summer of 2007.

The slides are well worth reading whether you are new to these topics or consider yourself an old pro. They cover introductory topics in distributed systems, motivate and describe MapReduce, and then discuss how to implement the PageRank algorithm and a variant on K-means clustering in parallel on a large cluster using MapReduce.

Wikia search: more insider details

Someone from the Wikia team finally came up for air on the Lucene mailing lists. It was Dennis Kubes, and aside from being really busy (as one might expect :), it sounds like they're trying to do the right things for the open-source community: publish their improvements back to the community, etc.

lucene-java: Dennis Kubes, "Re: Wikia search goes live today" (Nabble archive)

An interesting detail:

"Also the social networking functionality, named foowi, will become its own open source project (probably with an apache license)"

"foowi", I like it! :)

Whilst searching Nabble for this to share, I happened upon an earlier message by Dennis asking about Wiki search, "Is anybody here working with that project or planning on working with it?". I can only assume that Jimbo Wales pinged Dennis personally sometime after Mar 25, 2007. :)

nutch-user: Dennis Kubes, "Wikia Search Engine? Anyone working on it?"

Wikia Search finally launched... what is it?

The top news in the search world today is that Jimmy Wales' much-hyped "Google killer", Wikia Search, just pulled back the covers today for the world to see:

http://alpha.search.wikia.com/

Techcrunch called it a "complete letdown". That may be a bit harsh, but he's right, it's anything but a "Google killer" in its current form. Other writeups that I noticed include VentureBeat, Silicon Alley Insider, and John Battelle's Searchblog

So after a year of hype, what has Wikia actually built?

• It's clearly built on Apache Nutch and Lucene
• Click on a score link and the Nutch logo is still displayed, as of today
• It turns out that one of the Nutch committers, Dennis Kubes, has been (quietly) working on this project

Alongside the Google-style results are boxes for:

• "People Matching term" - user profiles with pictures
• "Mini Article About term" - links to a MediaWiki instance at search.wikia.com

In comments on the TechCrunch post, Jimmy Wales stressed that "It’s a project to build a search engine, not a search engine". I for one am glad they "released early", and can't wait to see where they're going with this. If they can actually make the "transparent index" that gives good results, it'll be a great service to internet users... more or less what Doug Cutting set out to do when he started the Nutch project.

(Disclosure: We used Nutch at my startup BusyTonight for all of our crawl infrastructure from 2005 to 2007.)