January 12, 2005

Giving birth to a new Internet

For those unaware of the effort, I thought I’d point out the existence of an interesting research facility called PlanetLab. In a sense it’s the next generation of clustering or grid technology — the ability to harness overlays of computing power to create virtual distributed computing resources. Somewhat confusingly the name PlanetLabis used to refer to all three of the following:

  • the overall project they are running,
  • the technology they are developing, and
  • a few hundred servers scattered around the world running some neat virtualisation software.

There’s a stupid network spin to this. Instead of the stupid network connecting smart devices, it ends up connecting smart virtual devices. What that means is that the next wave of decentralisation is being hatched.

The Internet grew from the ARPANET et al. Planetlab is to Internet Mk2 as ARPANET was to Internet Mk1. It’s an embryonic version of a future phenomenon. Let’s say PlanetLab is commercialised, and also grows like the Mk1 Internet. Taken to an the same extreme, you can imagine there wouldn’t be data centres filled with expensive fire supression systems and redundant power supplies. You just smear your data and processing across the virtual computing environment and let the natural redundancy of the Internet deal with the consequences of danger and disaster.

So if you’re looking for futures giants of data storage or application serving, it’ll probably come out of an initiative very much like this. Instead of hording data processing or computing power behind the corporate moat, the next iteration will turn that world inside out — the power will truly be at the edges. Some of the winners of the first waves of client decentralisation (Sun, Oracle, BEA, etc.) are likely to be hurt by this wave of server decentralisation. Probably too early to rejig your portfolio (and too late if you hold Sun stock). But one to watch.

Imagine if every corporation you dealt with had to store your data in a virtual data store under your own control? I look forward to the day!

UPDATE: Take a look at this excellent summary of the state of play in distributed content distribution. We really are in the midst of a huge technical and social change. Three massive shifts, one after another — many-to-ones (the Web), one-to-one (VoIP) and one-to-many (P2P). Each displacing whole industries in the process, and creating new ones previously unimagined. All enabled by Stupid networking. What a time to be alive!

Posted by Martin Geddes at 10:31 PM
Trackback Pings

TrackBack URL for this entry:
http://www.telepocalypse.net/cgi-sys/cgiwrap/mgeddes/MT/mt-tb.cgi/372.

Comments

i dig into it a couple of months ago but as far as i can see, it is just an overlay network over the internet. it is something many has tried before: napster, kazza and even bittorrent are some sort of overlay network.

and in the world of internet2, the overlay network that rules is DC so they are really pretty new. moreover, there seem to be more hype and promises in planetlabs (see comment by andy from intel) then the results.

it remains to be seen where this is heading but i wont be too quick to jump to conclusion.

Posted by: at January 13, 2005 02:47 AM

Perhaps the yardstick for "success" for PlanetLab is too long? To me, even if all we learn is what the difficult problems are then it's achieved its mission.

I suspect the stunted progress comes from the misbelief that the problem is the technical issues of clustering and virtualisation, rather than the economic ones of making it worthwhile for someone to participate in your network.

I can imagine the SETIs, BitTorrents etc. coalescing to form the actual de facto distributed computing platform. But PlatetLab at least points the way, if only in ambition.

Posted by: at January 13, 2005 09:41 AM

But they proclaim to be the "Next Internet", my yardstick is going to be *very* long :-)

I am also interested in virtualization & clustering - we have several virtualization projects going on now - blades obviously, vm, san, but also clustering xbox'es and now, i am going to do mac minis ;-)

but thats not what excite me abt Planetlabs - virtualization is done in far greater depth in grid forum like GGF and LCG. the promises of planetlab is a horizontal common platform for overlay network so we can stop trying to build more vertical overlay protocol for just one purpose. if it is successful, it could potentially as big as TCP...but thats a big *IF* which explains my yardstick ;-)

Posted by: at January 13, 2005 10:33 AM

I think James is right and the article is a bit too thick on the hyperbole -- caveat lector! One to watch with interst. Martin.

Posted by: at January 13, 2005 11:09 AM
Please enter your comment below. Your comment will not appear immediately -- they all go for pre-approval by me because of the volume of spam I receive.







Remember personal info?