In February 2016, I undertook an exercise to copy as many as made sense, of my original blog articles across from the oracle site to my/this wordpress blog. This article represents the highlights of the original record of my day.  The Oracle blog has now gone, as have the all the pointers to Sun resources, including the presentations. I rescued and rehosted Andy Ingram’s, Workload based Systems Design 2005 which I have rescued and reposted because it was important then and remains so today, well maybe, maybe not in 2019.

Sun finally launched it’s chip multi-threading systems, promising a revolution in throughput and cheaper MIPS/Watt. This was done at a synchronous event in New York & London with a webcast for those who couldn’t make it in person.  Jonathan Schwarz travelled to London to speak to his European customers, as did  I. I recorded this on my sun oracle blog in several articles. 

Niagara_Falls_w650

My blog, kicked off with an article called, Tomorrow’s Data Centre. It seems that 3m users/week are joining the net, they want different services and the data centre is the pain point between the staggering growth in demand for ‘net services and the environmental and management problems this causes. The speaker suggested that 60% of power growth is based on IT to meet the demand of power to compute and the power to cool. Sun plan to change the rules and change the capability/power equation.

I do wonder if being cleverer in the systems design side means that the horrendously inefficient management practices will continue, and whether we’re merely enabling crapness in service management.  particularly in performance management. It seems that what we need is a self healing performance management solution. This will/may come from the grid technologies, be it the grid engine, or developments of RIO or the use of virtualisation technologies.

The warm up act was followed by Jonathan’s show. He started with some illustrations of the changing nature of network technologies, the accelerating speed of change, the growth of reliability, and the transitions to service propositions i.e. cloud computing. Jonathan also very graphically, by showing the picture taken in the London tube on July 7th, illustrated the change taking place in the internet by transforming to a read/write internet, to creating value through co-operation. He then argued that openness and the sharing of intellectual property creates opportunity. It neither kills, nor expropriates it exclusively for the inventor. He tipped a nod to the workload profiling ideology (see me here …, and the corporate pitch here…). which should be at the centre of data centre design today. He then handed over to Mark Tremblay, the T1’s Chief Architect.

I originally blogged Tremblay’s presentation in an article entitled, UltraSPARC T1, throughput computing on my sun/oracle blog. Mark spoke of his design goals and then itemised his success in meeting them. He stated that in a world where competitors were offering 30% performance improvements; Sun’s T1 CPU based systems will offer integer improvements. (This is just as well; the US T1 has relatively poor floating point and struggles with fractions. This is a very technical and poor joke – you had to be there. 😳 )  Sun are shipping SPARC systems that offer 32 way SMP (8 cores/4 threads each) in a one rack unit, the CPU draws 70 Watts and the chip has 4 memory controllers on the chip. This has been designed to leverage the growing disparity between fast CPU and slow memory speeds, and to win business on the basis that faster can be cooler. The presentation had a slide with real heatmaps of the T1 system board and one of its traditional competitors.

For more on the next presentation, see Why Cool threads!, Mark handed over to Fred DeSantis, who presented a new metric for comparing systems, its called SWAP and is calculated by taking a capability outputs and dividing it by the product of space and power.It reminded me of some work I did for one of the bank’s two years previously, where I plotted two axis, Performance/Rack Unit and Heat/Rack Unit. Fred then called Mark back to the podium and they ran a data centre simulator,which provoked the thought in my mind that in the UK, we rarely model the A/C space & capital costs when exploring consolidation future state scenarios.

Two users were then asked to give testimonials, the first was Rene Weinholtz, from Strato AG, a web hosting company. You should have seen Jonathan’s face when he said that they didn’t have an environmental problem. Brits don’t expect to meet funny Germans. I’m a bit more broad minded since watching Deutsche Welle on Sky and visiting Berlin over the summer. Rene was actually very funny, but got serious when he described his system consumption in tonnes and stated that they expected US T1 to save them 85-90% of their space and power consumption. Mattias Schorer, from Fiducia AG stated that his company had seen 4-6 performance increase in their banking application and had in fact broken their load generator capacity. They had had to buy additional systems to complete their testing.

The final presentation I covered in an article entitled, a throughput ecosystem, We returned to Jonathan, who spoke of ISVs and services.  These would be key if his new business model is to be successful. He spoke of Sun Grid as Sun’s first IaaS offering and finished with the bombshell, that Sun is open sourcing the T1 design. The last word was left to (a video of) Tim O’Reilly, “Sun always was an open source company”.

I noted that the stock price went up after the launch.

ooOOOoo

Originally posted on my sun/oracle blog, republished here in Feb 2016, I have added the picture, and fixed the link to Andy Ingram’s throughput presentation as it remains relevant today. Another snippet I have carried over is,

I’d never heard that the telephone network switch was invented by a mortician who believed that his business rival’s girlfriend, the telephone company’s switch operator was putting business his rival’s way and wanted to remove her from the switch room and thus invented an automated switch.

Throughput Computing
Tagged on:                                                                 

One thought on “Throughput Computing

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: