It never rains but it pours. In the last week, there have been a whole lot of things happening in the tech world that I plan to write about, but if I write about them all at once then the posts compete with one another and if I spread them out then the posts lose currency. Before I pick one subject and dive into it, here’s a quick rundown of recent developments.

  • Steve Todd has been writing about Atmos, previously known as Maui, which is being billed as “Cloud Optimized Storage” by EMC.
  • Bryan Cantrill has been blogging about Fishworks, which is Sun’s latest attempt to get out of the storage-industry cellar.
  • Violin announced a 4TB flash memory appliance which – contra Wes Felter – is distinguished from Fusion primarily by access method and not size or color. Fusion’s differentiator is that they’re right on the PCI(-E) bus, not on the far side of some external interconnect. That’s actually pretty important, even though it would be far more useful if they hadn’t screwed things up by making their implementation specific to one CPU architecture.
  • ORNL announced the new version of Jaguar, the second machine to break the petaflop barrier and probably the world’s fastest computer (SC’08 is next week so we’ll see what the Top500 folks say).
  • SiCortex announced the Green Computing Performance Index to address shortcomings in the Green500, based on Matt Reilly’s proposal.

I plan to write separate posts about Atmos and Fishworks soon, but I’ll just leave them be for now. Hey Steve, since you’re making a series out of this, how about a post on how Atmos deals with global-scale latency and connectivity issues, so I can offer my own counterpoint on that near-and-dear subject? I’m going to focus on Jaguar and GCPI below the fold.

The funny thing about the Jaguar announcement is that it has been so low-key compared to the massive wave of reports about Roadrunner. Yes, I know Roadrunner was the first to break the petaflop barrier, but in some ways Jaguar represents a bigger leap. For one thing, it’s more powerful than Roadrunner not just on embarrassingly-parallel Linpack but on a host of other axes – especially I/O – as well. It’s also the first non-IBM system since the advent of Blue Gene to capture the top spot. According to, the Livermore BG/L system captured the top spot in November 2004, and held it until stablemate Roadrunner took the lead in June of this year. Besides the shift from IBM to Cray, this also represents a shift from west to east, and a bit of a retreat from parallel increases in power and efficiency. Those last two are not mere coincidence. I guess you can make it big without making it efficient after all, if you happen to have several hydro dams nearby to power your machine – exactly why Jaguar, along with successors Kraken and Baker, is at ORNL instead of somewhere else.

This, obviously, brings us to SiCortex’s Green Computing Performance Index. Unlike certain other vendors’ attempts to gain competitive advantage by dismissing current benchmarks in favor of a non-existent and ill-defined “benchmark of the future” they actually have no corporate interest in seeing become real, GCPI is a concrete proposal being made today, with full disclosure of methodology and underlying numbers. Yes, the results that are there today do make the SiCortex systems look good. That’s because we’ve had our eye on this particular ball more than competitors have. However, anyone can contribute their own numbers, or rearrange the numbers that are there to create a “single figure of merit” that they think is more useful, and we’re actively seeking a less conflicted host/sponsor for the database. More importantly than who got this ball rolling or who “wins” according to the current numbers, we all win when benchmarking is done right. If “green computing” is to be more than just empty claims, we need quantitative measures of how green computing solutions really are, and the GCPI is a constructive step toward that.