This week we’re stepping back from our usual 7-day shot of what’s going on in the world of HPC and taking a peek at the long view. I’ve combed through the insideHPC.com archives to come up with some of the most popular themes and news items from the full year of coverage, as determined by reader page views and clicks.

First, some stats. We started publication of insideHPC.com on Dec. 28, 2006, with post one. As of this writing (early morning on Dec. 20, 2007), we have written 1,273 posts about HPC news. Our readership has grown too, from 1 (me!) on Day 1 to nearly 1,000 visitors a day in December of this year, and all from the very tightly knit world of HPC.

There are two biases in this analysis. First, about 40 percent of my traffic comes from search engines, and that has grown over the year. These referrals are to deep links in the site, and this traffic biases in favor of posts that have been around longer. insideHPC.com has a fairly high page rank, which means that we show up on the first page for lots of searches in HPC. In fact, many of the same searches for which HPCwire shows up (a good thing!).

Second, readership has built steadily throughout the year. As more and more readers turn to insideHPC for a daily swig of HPC news, the pool of potential readers and clickers grows, which biases in favor of the more recent posts.

I’ve made absolutely no attempt to correct for these biases. But, I was surprised to see posts from as recently as this month alongside posts from March show up in the site’s top content. So it may be that this summary is at least partially representative of what the HPC community thought was interesting throughout the year.

Without further ado, and with so many caveats you’re going to need an extra Barney’s bag to tote them around in, I give you:

The insideHPC Top 5 Stories for 2007

>>HP and Bull getting hitched?

As it turns out, no. At least not in 2007. But back in late July rumors were swirling that the two companies might get hitched, providing HP a native hook into the European HPC market, and a strong local service presence. The deal was thought to be worth just shy of $1B US at the time, and the story was covered in major tech news outlets across the country.

We can cross our fingers that they kept each other’s numbers in speed dial though, and that they might finally get together in 2008. I mean, Rachel and Ross got back together, right?

http://insidehpc.com/2007/07/26/rumor-hp-to-buy-bull/ (And the related news links on that page.)

>>NVIDIA’s magic cards

NVIDIA was much on the minds of our readers as the company moved to build a new market for its legendary graphics cards. They aimed at the HPC market not only with some tweaked GPUs, but they also took a page from ClearSpeed’s playbook by introducing an API (CUDA) to allow HPC users to get at all that processing power without contorting scientific algorithms into constructs better suited for textures and pixel shading.

Top NVIDIA links: http://insidehpc.com/2007/07/17/nvidia-demos-cuda-plug-in-for-matlab/, http://insidehpc.com/2007/06/21/nvidias-tesla-2-tflops-on-your-desk/

>>CFS, ZFS, Sun, and Lustre

A story that seemed to pique everyone’s interest this year points directly at a big pain point for those of us that run large HPC centers: filesystems. Sun and Cluster File Systems, a commercial provider of Lustre, started making news back in the summer when CFS announced that they would be welding Sun’s ZFS file system in as the back end for Lustre on all platforms, not just Sparcs.

Then, early in the fall, the other shoe dropped: Sun opened up its wallet and bought CFS lock, stock, and filesystem. The potential negative impact of this acquisition was large, as Lustre is a filesystem of choice for customers on just about all the major HPC platforms. Sun pledged at the time to keep Lustre available on competing platforms, and so far has kept its word.

http://insidehpc.com/2007/07/14/cfs-moves-lustre-to-zfs/
http://insidehpc.com/2007/09/12/the-sun-sets-on-cfs/

>>Google does an HPC grab and dash

Computing behemoth Google is always just on the periphery of discussions in high performance technical computing. As we plan our 20 MW power plant upgrades and scrape together $35M to buy a new system every couple years, we marvel at the scale and efficiency of what Google has managed to build. Or at least what we think Google has managed to build: the company is notoriously tight-lipped when it comes to hard details on the main source of its competitive edge.

This year Google reached out and touched us directly, however, when they bought parallel code development startup PeakStream.

Everyone agrees that the coming multicore chips are not going to yield their performance gains to the standard SMP programming model, and we’re going to need new tools to make good use of supercomputers in the very near future. At the time of the acquisition, there were two companies leading the charge to develop these tools: RapidMind and PeakStream. After the acquisition, Google let it be known that it had every intention of keeping PeakStream’s IP to itself, which left us with one tool developer. Let’s hope they get it right.

http://insidehpc.com/2007/06/06/peakstream-mystery-check-googles-left-pocket/
http://insidehpc.com/2007/06/07/analysis-on-the-peakstream-google-acquisition/

>>AMD, AMD, AMD

AMD’s year can be summarized in three letters: WTF.

From plummeting stock prices to missed delivery schedules and chips that don’t work, AMD had the kind of year that hurls top management on the refuse heap of corporate America. Of course there has been plenty of this kind of luck to go around, and in years past it’s been Intel’s turn in the stockade. But this year the tables turned, and Intel seemed to do everything right even as AMD did just about everything wrong.

Most recently of course we’ve learned that AMD’s bragging throughout the year on the ill-fated Barcelona was a bad…no, really bad…idea. Originally scheduled for mid-2007, the chip wasn’t ready until near the end of the year. The delay added insult to financial injury when SPEC invalidated all Barcelona benchmark results for lack of general availability. And then, just when AMD thought it was ready to finally go forth and ship product, a nasty bug turns up in its TLB silicon, torpedoing performance and jeopardizing a key market for the company.

It remains to be seen how this bug will impact HPC. Early Cray XT4 shipments that use the chip appear to be on schedule at this time because HPC workloads appear to not stimulate the bug. I’m skeptical until I see it run in daily production, however. If I had a dollar for all the times that something that I was told “shouldn’t make any difference” grounded my HPC center, I’d be a rich man.

The AMD posts in the top 25 for 2007 are at http://insidehpc.com/2007/12/06/amd-delays-barcelona-ramp-up-to-q1-2008?referer=sphere_related_content and http://insidehpc.com/2007/12/06/more-on-amds-tlb-bug-and-damaged-honor/. But a truly excellent 2007 retrospective of AMD’s Year of Tears can be found inside HPCwire at http://www.hpcwire.com/hpc/1947254.html.

From all of us in the happy band of news hunters and gatherers at insideHPC.com, Merry Christmas!

—–