IO500 Team Visit

IO500 Benchmark Gets Traction

Storage is complicated and benchmarking it has too many complexities for the traditional kernel-like or application-specific approaches. Thanks to a few experienced and tenacious researchers, and the community that supports them, the IO500 has managed to put a credible stake in the ground, and is getting traction, with 101 entries on the current list and expecting many more by SC19.

ReadioFreeHPC hosts the IO500 Steering Committee to do a deep dive. "The steering committee is the decision body ensuring the development and curation of the benchmark and its results but also responsible to resolve ethical issues." Henry and Shahin ask the hard questions, or so they think!

John Bent (Seagate), Julian Kunkel (University of Reading), and George Markomanolis (Oak Ridge National Laboratory) join RadioFreeHPC's virtual studio. We missed the fourth member of the team, Jay Lofstead (Sandia National Laboratories) due to scheduling conflicts. We also missed Dan once more as he was navigating airports and planes coming back from Australia.

"The IO-500 has been developed together with the community and its development is still ongoing. The benchmark is essentially a benchmark suite bundled with execution rules. It harnesses existing and trusted open source benchmarks. The goal for the benchmark is to capture user-experienced performance.

Henry Newman's Feel-Good Security Corner

The segment that is rapidly establishing itself as the go-to place for why being online is just too dangerous. Our spirits are lifted again as Henry describes a ransomware attack on a back-up site for dental offices in Wisconsin. There go insurance data, contact information, etc.

Ransomware Bites Dental Data Backup Firm

PerCSoft, a Wisconsin-based company that manages a remote data backup service relied upon by hundreds of dental offices across the country, is struggling to restore access to client systems after falling victim to a ransomware attack.

Catch of the Week

Mining cryptocurrencies is compute intensive. The high levels of required electricity has made the topic visible and controversial. So where would you go if you want a lot of electricity? Why, the nearest nuclear power reactor, of course. Shahin talks about crafty folks who have done just that!

Employees connect nuclear plant to the internet so they can mine cryptocurrency

Ukrainian authorities are investigating a potential security breach at a local nuclear power plant after employees connected parts of its internal network to the internet so they could mine cryptocurrency.

Henry describes a few ISPs who ended up stealing communication spectrum from, guess where, the airport, obviously. And here's the thing about at least some of these incidents: the whole thing is so complex now that it can be hard to tell incompetence from malice.

American ISPs fined $75,000 for fuzzing airport's weather radar by stealing spectrum

Three ISPs will be fined $25,000 apiece by America's broadband watchdog, the FCC, for interfering with weather signals in Puerto Rico. Boom Solutions, Integra Wireless, and WinPR were all found to be using devices for their point-to-point broadband that were “misconfigured,” according to the regulator this week. This caused interference with a doppler weather radar station at San Juan international airport.

Listen in to hear the full conversation.

Download the MP3 * Subscribe on iTunes * RSS Feed

Sign up for our insideHPC Newsletter

Forty+ different AI chips

What are we going to do with 40+ different AI chips?

This week, the team looks at AI chips again, this time motivated by an article in EE Times about once such chip, Graphcore, and touts it as "the most complex processor" ever at some 20 billion transistors. The VC-backed company out of Bristol, UK is also valued on paper at $1.7b, gaining it the coveted "unicorn" status, apparently the "only western semi-conductor unicorn".

This being one of 40+ such AI chips (and that may be conservative), the odds of success are tough and the task formidable. But even if only 2 or 3 of such chips become successful, that's already a significant disruption to the market.

The Graphcore chip is 16nm, 1.6GHz, and comes in a PCIe card at 300W. You can stack 8 of these in a 4U chassis, so 2.4 kW just for those.

After a mini-rant about respected publications succumbing to clickbaits, the team talks about how cooling will be an issue and calls again for more clarity in performance metrics since the chip is rated at 125 TFlops but we don't know at what precision. Shahin reminds the team of his suggestion to clarify things by including precision in the metric, like DFlops for double precision, and then S for single, H for half, and Q for quarter precision.

Henry talks about how hard it is to build and test complex software like this despite Shahin's view that the modern software stack is too high so the chip need only be concerned with a couple of layers, codes are new and open to getting recompiled, it's increasingly open source, cloud providers and large customers have the wherewithal to do the job, and traditional HPC customers have the willingness to do the work if performance enhancements are there.

No "Catch of the Week" this time since Henry had a hard stop. We're used to it!

Listen in to hear the full conversation.

Download the MP3 * Subscribe on iTunes * RSS Feed

Sign up for our insideHPC Newsletter