Comments on: A bandwidth breakthrough

By: Bri

Bri — Fri, 26 Oct 2012 19:31:27 +0000

We are a nation of sheep, and we are getting fleeced. Right now we have mostly scabby skin.

By: Jim Mooney

Jim Mooney — Fri, 26 Oct 2012 18:53:50 +0000

Gee, maybe this means that someday our average internet speed may approach half that of Korea. But it will still cost an arm and a leg compared to most of the civilized world. It’s so sad that we invented the net but are now lagging way back in speed, yet paying much, much more for the rotten service. And even though a better internet is known to boost the economy when we sure could use that.

There is something seriously wrong with out whole system. The truth about its massive flaws is there, but it’s just not getting out.

By: Andrew

Andrew — Wed, 24 Oct 2012 15:12:05 +0000

Sounds nice, as long as the information sent can easily be described by some linear equations. That kind of compression probably doesn’t work for all data types, though. Encrypted information might be especially difficult to model.

By: Chris P. Kareem

Chris P. Kareem — Wed, 24 Oct 2012 03:48:27 +0000

This is a good improvement over current technology, but I don’t see why TCP doesn’t just use asynchronous packet transmission. As far as I understand the problem is that packets are sent consecutively, and when one gets lost there is a delay in the flow of bits because the router has to re-request the packet before anything else is sent. But why not just send the rest of the packets out of order, and while the rest of the information is being sent, the router can just re-request the lost packet in the meantime? That would be more effective without the overhead of the methods in this post, unless I’m mistaken as to what the problem is.

By: asiwel

asiwel — Tue, 23 Oct 2012 17:46:20 +0000

This looks like a reasonable method of “compression” if it works out well under a wide variety of conditions. I more or less take for granted that the researchers are competent, that MIT would not waste time testing something that had obvious flaws and weaknesses, that peer review would have detected reinventing the wheel, etc.

By: GAUSS

GAUSS — Tue, 23 Oct 2012 15:20:47 +0000

Look up the proposed “IPComp” standard from 2001.

By: Marcos Marin

Marcos Marin — Tue, 23 Oct 2012 15:03:50 +0000

- No, because you are not solving the problem. What happens if the packets describing your “smaller” program get lost?

- Not necessarily. This is the realm of Information Theory (though they misinterpret their results and can’t see the whole picture). In this case, however, yes! But this is not the point, the point is, does this “extra data” is significantly less than the extra data you are already sending when packets do get lost?

By: Marcos Marin

Marcos Marin — Tue, 23 Oct 2012 14:53:41 +0000

They are probably referring to techniques such as used by software akin to “quickpar”, which are so old one has to wonder why network engineers only now come up with this.

If you allow me a simplification, it’s just like we did in school when solving for x. If x+3=5, then x is 2, now each packet will have its own equation and variable, x, y, etc…

Maybe someone can follow the link there and confirm this is the way they are doing it, if not, well they should =)

By: SamCrawford

SamCrawford — Tue, 23 Oct 2012 13:26:40 +0000

Ian, I’m guessing that the method used is similar to the Parity Files method of rebuilding rar files.

http://en.wikipedia.org/wiki/Parity_file

By: SamCrawford

SamCrawford — Tue, 23 Oct 2012 13:24:26 +0000

This has potentially huge implications. I for one am very excited by this prospect. And 2016 is an extremely short timescale for these changes to take effect. A 25 fold increase without increasing spectrum is huge.

This is a huge leap forward. I’m sure we’ll be hearing a great deal more about this in the very near future.

By: Ian Clarke

Ian Clarke — Tue, 23 Oct 2012 13:17:22 +0000

“The technology transforms the way packets of data are sent. Instead of sending packets, it sends algebraic equations that describe series of packets. So if a packet goes missing, instead of asking the network to resend it, the receiving device can solve for the missing one itself. ”

How can packets go missing if it’s sending algebraic equations instead of the packets? Are they referring to missing packets of equations that describe the original packets?

Good news for our overloaded networks tho’!

By: MarkOates

MarkOates — Tue, 23 Oct 2012 12:45:19 +0000

If I understand correctly, basically they’re compressing the datastream. In this case, the packets themselves are being ‘compressed’ by representing a packet with an algorithm that describes it.
Makes sense to me. Instead of sending assembled code, send a smaller program that assembles it.

It seems like the downside is that more information is being delivered. If no packets are lost in the transfer, wouldn’t the extra algorithm just mean more data?