A Decade of Wonder
By marchamilton on Dec 31, 2009
I told a friend of mine yesterday that I was taking the holidays off from blogging but couldn't resist an end-of-year blog.
Anyone in IT a decade ago surely remembers the attention paid to the passing of "Y2K". One Sun employee even wrote a song about it. And on the scientific side of computing, Sandia National Labs was celebrating their recent #1 entry in the Top500 list with the famed ASCI Red supercomputer, reaching a whopping 2.3 TF using 9632 processors.
But as I told my friend, I've been spending the holiday break training for the upcoming LA Marathon and catching up on my holiday reading, courtesy of my wife and two kids. So no surprise, one of the books I read was Born to Run. I have to thank Borje for recommending this book to me last month, but it wasn't until I gave it as a gift to my son, a recent convert to cross-country running, that I took the time to read it myself. Thanks Borje and Evan.
Of course the passing of a decade can't go without some nostalgia, so thanks to my wife I went back several decades to the Apollo program and read Rocket Men. Definitely brought back memories of my days at TRW in the 1980's. As a new hire at TRW, you were indoctrinated into the company's history in the Apollo program, having built the world's first throttable rocket engine used as the lunar lander decent engine. Our new hire film also showed scene after scene of earlier failed rocket launches, reminders of what could happen when things went wrong. While Rocket Men did not discuss it, one of my favorite tidbits of Apollo program history was the role of the TRW-built backup navigation system, a so-called strap down inertial navigation system, in safely bringing back the Apollo 13 spacecraft after its near-devestating accident damaged the primary navigation system and forced the craft to limp-home on minimal power. I guess that strap-down system was the world's first green navigation system.
Last but not least in my holiday reading, a step back to an even earlier day in science, The Age of Wonder, when science and art came together in wonderful ways.
So what will the coming decade bring? As we close this decade, Sandia is still the Top10 of the Top500 list, this time with a Sun Constellation System, aptly named Red Sky (and also one of my most frequently read blog entries for 2009). The Red Sky system, at 423.9 TF, is about 185 times more powerful than the decade-earlier ASCI Red system and in fact it would take only about six Sun Constellation System blades, about 1/8th of a standard 19" computer rack on Red Sky, to equal the total compute power of ASCI Red. While we don't know who will be at the top of the list in 2019, we do expect it won't be a TeraFlop or even at PetaFlop system but in fact an ExaFlop system.
Another topic sure to be top of mind in the coming decade is cloud computing. At the hardware level, many of the architectural concepts of cloud computing were actually first developed in the world of HPC and Top500 system. Quick, without thinking too much, tell me if the picture at the top of this blog is of a leading cloud computing data center or a leading HPC data center? If you recognized the picture as the TACC Ranger system (the #9 entry in the latest Top500 list, and like Sandia's Red Sky, also a Sun Constellation System), you'd be correct. But with 1000's of identical servers connected with a high performance network and multi-petabyte global file system, it also makes a perfect cloud data center. As our soon to be new CEO likes to point out in interviews, the basic hardware concepts of cloud computing, a server connected to the Internet, are not new. In fact, Oracle was one of the pioneers in cloud computing with their earlier Network Computer division. But what IT topic wouldn't be complete without a Bruce Kerr song.
Over the next decade, the computing industry will continue to be challenged to build larger and faster computers. But by and large, the hardware paths to the ExaScale systems of 2019 are well on their way to being designed at semiconductor companies around the world. Many of the greatest computing challenges of the coming decade are likely to be software challenges. For while the peak computing power of microprocessors, servers, and in fact entire HPC or Cloud data centers continues to grow according to Moore's law, getting software to scale is increasingly the greater challenge. Solving those challenges will require ever closer cooperation between hardware architects and software architects. Those in the industry who simplify the coupling of software with hardware are likely to be winners in both HPC and Cloud Computing. I really look forward to blogging about how we solved those challenges in 2019.