Bitwrangler
Neuroinformatics in Tulsa
A couple of weeks back I was fortunate enough to be asked to speak to a group of Neuro Scientists and Computer Scientists at a Neuroinformatics conference that is partially sponsored by my Alma mater. It is difficult to tell from a one day snapshot where this will all go, but the science seems very interesting and they have formed the nucleus of a very good team of partner schools and scientists. It is something I'll be watching for the next few years as it blossoms.
More information on the conference can be found here: http://www.ibcb.utulsa.edu/
Posted at 07:58AM Apr 27, 2009 in Personal | Comments[0]
Michael Dell at Supercomputing 2008 Huh?
Unleashing Human Capability
The 4th wave of HPC
Michael touched on the ubiquitous nature of HPC. He mentioned the CERN installation, claims it will need 20PF to duplicate human brain function. It would cost $3.6B and would not match the power utilization of the Human brain. Then he starts pandering to our ability to solve big problems. We know that, thanks. Points out Moore's law stuff. Again, thanks.
1st wave Specialized Microprocessors
2nd wave Microprocessor based parallel systems (thinking machines) and started into a laptop sales pitch then the GPU sales pitch.
3rd wave Standard Clusters -- Says Dell is famous for this.. (really?) Who knew it only started with Dell in 1999? Huh.
4th wave: High density, better management, lower energy, etc. Now we are fully into a Dell product pitch.
Speeds, feeds, simple arithmetic. Awesome.
Claims they will be first to market with QDR IB. Who's switches?
And now (at 9:30) the crowd is starting to walk out shaking their heads. Shocking.
And then followed the first URL reference. Excellent.
Reference to facebook. Hmm. Very questionable reference. BP, at least they are doing HPC.
I find myself unable to look away as the train wreck unfolds before my eyes.
Now we have the reference to Hyperion. Thank you Mark Sieger.
It is official, we now have done worse than Bill Gates as a SC Keynote.
Posted at 07:37AM Nov 18, 2008 in Sun | Comments[0]
Solaris and Easy now fit in the same sentence
Over the 4th of July holiday here in the Great State of Texas, among other celebratory events, I decided to re-organize the data center at the dam house (not a pejorative term mind you, we live below a hydro-electric dam). During the cleaning, backing up, organizing and migration two interesting events were undertaken.
First, I migrated from an old AMD 32 bit shuttle box to a AMD 64 socket M2 system (ASUS MOBO, 2 Gig of ram, .5 TB hard drive) with my Microsoft Windows XP system. The overall process to install XP on the system went like this:
- Load CD and start the boot from CD process.
- At checking hardware.... black screen never comes back.
- Check BIOS settings repeat step 1&2 several times with intermittent head scratching.
- Pull hard drive, attach to another machine with linux image, run fdisk wipe out the existing linux partition on the disk.
- Replace hard drive, and re-start with step 1
- After OS load, use MB pro to gather all drivers from ASUS web site for on-board chipset and nVidia web site for graphics card burn CD and use it to finish the install.
- Attempt to move "files and settings" from old machine to new machine, experience 3 failures and 1 successful completion.
- Re-install all applications on the new machine.
After getting the XP (legacy machine) up and alive and after hearing horror stories of lost data from a couple of friends who had recently lost hard drives on various machines I decided to make a bold addition to the datacenter in the form of another set of hardware that was identical to the XP machine in every possible way that happened to be available in the bitwrangler overstocked warehouse, dive shop and coffee emporium. On this kit I decided to give OpenSolaris a go so that I could build a ZFSed NFS server for the local backup of important data, pictures of the two cherubs and mom (bittergirl). Sounds like a good thing for a Sun engineer to do and it should be well within my capabilities. (Afterall, I architect HPC systems for a day job.)
It had been a solid year since the last time I installed Solaris X86. At that time, I had been running it for a solid year on my ferrari laptop and had been quite satisfied with the overall user experience but expected a day of "futzing" to get everything where I wanted it. I burned the CD, loaded it, answered a couple of questions and voila, it came up with all of the drivers and attached to the network with no extra work required for a basic system. Another 5 minutes choosing services and the entire configuration was complete, up an running.
Total time (user, not wall clock) maybe 5 minutes. Compared to the XP experience (just to get to the running system) 3 hours or more. Outstanding!
CONGRATULATIONS! to the Open Solaris community. You've done a fantastic job to this point. Keep up the good work.
Posted at 12:07PM Jul 08, 2008 in Sun | Comments[0]
We have a petaflop
IBM with BlueGene/L has hit the petaflop at LANL. The mainstream story is here. IBM & LANL press releases are not out yet. Curious.
Anyway, it looks like the Petaflop Rex has been discovered. Nice bragging rights to be sure, but what does it mean in terms of real application capabilities? It will be interesting to watch the spin for the next few days.
Posted at 07:37PM Jun 08, 2008 in Sun | Comments[0]
Advanced Nutritional Research-- Long Live the Brisket
A masters thesis at Texas A&M by Stacey Turk has un-covered the heart healthy aspects of the fat found in beef brisket. More info is available here. Clearly this is a controversial enough finding in today's health conscious, eco friendly world that more study is definitely in order. To that end, I suggest the in-state rivals Texas A&M and The University of Texas need to open some joint research on the issue and apply some of the power of Ranger toward the protection of the primary food of the great state of Texas. Tofu is dead. Long live the brisket!
Posted at 11:20AM May 28, 2008 in Personal | Comments[0]
HPC Meets Scrapheap Challenge
HPCwire brought something back to the front of thought process today with their article on the SC2008 Cluster Challenge. The challenge is in it's second year and it is an attempt to demonstrate the "democratization of supercomputing" whatever that means. In essence, you take a group of students, a rack of gear and you run some codes in a timed event that starts with step 1) Un-crate the hardware. Except for the hardware coming from some sponsorship and being production level stuff, it struck me as a high tech version of the BBC Scrapheap Challenge (definitely not the horrid Americanized version Junk Yard Wars) program from the turn of the century.
For the un-initiated, in Scrapheap Challenge, two teams were given a task, 24 hours and a literal junk yard of resources to build the project. Each team had an expert in the field of study and any task specific supplies that wouldn't be normally found in the junk yard were usually hidden in the junk yard. They built sub-marines, salvage rigs capable of picking a car up off the bottom of a lake, rockets, punkin' chunkers, all sorts of geek engineering (mostly guy) fun. Mostly it illustrated the difference between mechanical engineering (we build projectiles and things that throw them) and civil engineering (we build targets). In a way that only the Brits can do really well, all of the desperation, mistakes and false starts were shown in detail and in the end, the result usually worked within reasonable levels of success. The US version was just a little too sanitized for my taste.
So, I think this cluster challenge could have some of those great earmarks without (hopefully) the sparks of SHC. It seems that there is a lot of preparation that went into year one, so a lot of the drama should be behind the teams. Murphy loves trade shows however and I suspect there will be un-expected events to follow.
I regret not having time to invest in the event during the Reno show last year. Frankly it got lost in the fine madness that is the Super Computing conference. I will attempt to follow it a little more closely this year because it just seems like good old fashioned geek fun. Of course there is this nagging little customer install that is scheduled to heat up about then....
So, stay tuned and keep the propeller on your cap spinning please.
Posted at 08:45PM May 27, 2008 in Sun | Comments[0]
More Ink on Ranger at TACC
The Sunday Austin American Statesman provided front page coverage for Ranger at TACC. (you can read the story here) For those of you at Sun who remember BitterGirl, you owe her for this one. It seems the Statesman was looking for local Tech stories and she sent them there for a story that would benefit her company. Lo and behold, they found an even bigger story with Ranger.
Who knew it was a secret?
Posted at 10:33AM May 27, 2008 in Sun | Comments[0]
I Smell Spin
It seems that the Itanium Solutions Alliance is pushing spin the likes of which we have not seen since Ross Perot ran for POTUS in 1992. Here, they are pushing Itanium as the fastest growing chip on the market. The immediate Rossism that came to my mind when I read that was from the last debate on Thursday before the 1992 Presidential Election when Ross responded to then Gov. Bill Clinton's assertion that the average Arkansas citizen had doubled his net worth during his tenure as Gov., by saying: (Invoke Ross Perot voice here) "Now folks I'm here to tell you that if you have a penny and someone gives you a penny then you have just doubled your net worth."
Similarly, Itanium has never had a large market share when compared to the X64/X86 architecture in general and with the X64/X86 market split into AMD/Intel and further divided into Opteron, Barcelona, Harpertown, Nehalem and a dozen more that are still on the market in one form or another and then you take into account the individual release schedules of both AMD and Intel you have a situation where a chip lifespan is well under a year. How long is it between Itanium releases?
Doubling your market share in at least 2x the life span is at best holding your own. Oh well as they say, there are lies, damn lies and statistics. The highest form of this art is Marketing.
Posted at 07:50AM May 21, 2008 in Sun | Comments[1]
Congratulations to One of our Own
John Gustafson, formerly of Sun Labs and Clearspeed landed as CEO at Massively Parallel Technologies (MPT). I've known John for around 18 years since I was a pup working for nCUBE. Congratulations John, I for one at Sun wish you well in your first gig as CEO. You can read more about this here.
Posted at 07:30AM May 21, 2008 in Sun | Comments[0]
ORNL upgrades Jaguar
The petascale rumor mill has been saying that LANL is likely to have the first Pflop machine installed later this summer. That machine is to be (so the rumor mill says) a big blue machine (or upgrade). It looks like the second place competition for DOE money was installed and announce last week with the upgrade to the Cray Jaguar machine. With this upgrade, Sun will put at least some $ in the coffers through the Lustre relationship with Cray.
You can read more about it here.
Posted at 09:50AM May 19, 2008 in Sun | Comments[0]
Microsoft Woes of the Future?
In a recent Gartner article on Microsoft Windows Broken Nature here they say:
"Microsoft's operating system (OS) development times are too long and they deliver limited innovation; their OSs provide an inconsistent experience between platforms, with significant compatibility issues; and other vendors are out-innovating Microsoft
. That gives enterprises unpredictable releases with limited value, management costs that are too high, and new releases that break too many apps and take too long to test and adopt. With end users bringing their own software solutions into the office...well, it's just a heck of a sad story for Microsoft. "
If you couple this with their big time push over the last 3 years or so to go into HPC (they presented at the Salishan Conference last week for goodness sake) and the HPC environment's desire to push past Petaflops and on to Exaflops by 2018 or so it paints a very scary potential future. One should never under estimate the influence of the 30Ton Gorilla in Redmond, but is this influence going to be good for genuine HPC? When we start talking about multiple thousands of nodes and millions to billions of concurrent threads, the OS becomes a big time issue. Between OS size and OS jitter, we already have significant trimming to do in the OS Kernel on the node. This is in direct opposition to the normal ways of Microsoft.
Scary. Very Scary.
Posted at 06:48AM Apr 30, 2008 in Sun | Comments[0]
The treadmill continues
NASA Goddard has released a draft of an RFP. Another plea for petaflops for picobucks. Sharpen your pencils, run your benchmarks and sell the future. Here we go!
Posted at 12:06PM Feb 29, 2008 in Sun | Comments[0]
Maybe they can research water polution clean up efforts
Repeat after me. WOOOOOOOO PIG SOUIE! The University of Arkansas (AKA the Pig People) have a new supercomputer courtesy of Dell. As a transplanted Okie, who graduated from a rival college back in the days of the Missouri Valley Conference I offer a hearty congratulations to the wearers of the orange pig and two suggestions for possible research: 1) Can you figure out what makes a NCAA Football fan put an orange pig on their head? And 2) How about searching for clean up techniques to take care of the chicken poop pollution from Arkansas chicken farms that Gov. Clinton allowed to flow into Oklahoma through the Illinois river (A condition that continues to this day.)?
Posted at 09:44AM Feb 20, 2008 in Sun | Comments[0]
Never Underestimate Ethernet
Woven systems announced HPL benchmark results on par with DDR IB. While I'll be the first to argue, with all due respect to top500.org, that all the world is not about HPL, I am convinced that one should never count the ethernet world out when it comes to price/performance metrics. I've been saying for a while that when they get 10Gbit ethernet NICs down to $30 a copy and 10 Gbit ethernet port costs down then they will give IB a serious run for the HPC network of choice (in volume). At the high end of the top500 there will always be special interconnects, but for the bottom 400,000 clusters, ethernet will be just fine thank you. You can find more about the Woven announcement here.
Posted at 09:34AM Feb 20, 2008 in Sun | Comments[0]