FEATURES & COMMENTARY
San Diego, CALIF. — This is an appendix to the recent article “SUPERCOMPUTING TAKES YET ANOTHER TURN.” (18939, 11.23.00)
Whilst the article did have an excellent balance of views, I would disagree with the title of this commentary, suggesting “NEW USES FOR THE INTERNET” a more appropriate title.
Supercomputing hardware’s eternal challenge is latency – the “wait time” from requiring data until it is available. As long as current mainstream physics remains valid, Einstein’s theories state that information takes at least a linear function of distance to travel between points in space. Using compute nodes through the Internet – a geographically dispersed entity – has the speed of light a limiting latency-generator.
Compounding the physics are the real-world problems of an Internet-compute farm of:
1) Loadbalancing. The many varied device that is the Personal Computer, has varying capacities and perfomance in areas of processor, memory, storage, network. Any problems requiring interaction in a time domain will run at a speed of the slowest node.
2) Security. It is true security can never be assured, the most significant threat is from the “inside” – those with access to the systems and data – in the Internet-compute-node model this access is distributed and provides more routes for hacking.
3) Robustness. The PC is a complex mix of devices which we all know to be a recipe for unreliability. Compound this by the unsecure home/office environment where these desktops reside, makes the likely failure rate of nodes relatively high.
4) Economics. There are a limited number of people who would trust remote use of their PC with limited/no financial return. This starting pool is going to be difficult to grow. Compounding this is the cost in power+cooling of an active PC (power save modes would disable in an active system). An energy expert could comment on the environmentally damaging aspect of the higher energy demands per work-done of a distributed model.
I also wish to comment on the future of IT and the link with HPC.
The PC has grown on the back of 2 uses, business and entertainment.
Business applications, which initially drove performance, is now largely met by relatively slower performance devices. For example, all my office applications run on an old 75mhz system, whilst I wait for my small 1Kg 300mhz laptop to be repaired. Both HPCwire, and this reply were read and entered on a PDA then emailed directly through a cellular handset.
Entertainment is a stationary event – to watch/engage suits being sat down. I see a merging of the TV, Radio, Video, High-Perfomance Games platfom, into a TV-replacement. As such, the inconsistency of the hardware in the future PC is unlikely to be abated.
People prefer mobility, and will seek out effective solutions.
Bring these threads together, and one can (possibly) predict a future of:
1) personal devices merging communication and information. A human has a high tolerance for latency – considering data in 0.01 seconds to be effectively “instant”. This suits a data query personal wireless model.
2) home entertainment solutions merging the personal devices with “richer” experiences such as video entertainment, here higher bandwidth can domestically be piped. The home environment is the “docking station” for the Personal Device. These systems, and the Personal devices can call on the land network for data.
3) Traditional “big iron” systems that include all traditional elements of HPC – perfomance (low latency, high bandwidth), robustness (redundancy, power continuity, e.g. UPS), capacities – so everything is available to all. These are managed in secure environments, which also sell HPC cycles to all.
The Internet does have the potential to suit independent problems with little/no interaction between nodes. but this is not Supercomputing – just a possibly additional use of the Internet.
True Supercomputing has an exciting future, with new computational solutions in physics, chemistry, biology, numerical engineering simulation and virtual prototyping, set to explode the size of HPC systems.
I agree the article’s observation that we lack the software, but through the Internet more people have access to larger systems to help overcome this difficulty.
— Nigel Healy Tel 07010704366, Fax 07010704367
The views experessed in this article are those of its author and are not necessarily endorsed by the publishers of HPCwire.