Fujitsu develops new data transfer protocol 30 times faster than TCP

[derpenxyne]

http://www.extremetech.com/computing/147152-fujitsu-develops-new-data-transfer-protocol-that-is-30-times-faster-than-tcp

[andrewcooke]

if the company divulged its secret, it wouldn’t be able to commercialize the new protocol

why? does it contain nothing new? have software patents suddenly become worthless?

[jpdoctor]

As usual, anything from extremetech is worth ignoring.

[notatoad]

What is the actual time associated with the TCP protocol? I'm imagining that 30x faster is pretty much an unnoticeable difference in real-world usage.

[wmf → notatoad]

They mean 30x throughput, not 30x less CPU time.

[notatoad → wmf]

I think if that were their claim, they might be running up against the laws of physics. Wires aren't going to carry 30x their current throughput no matter what the protocol is.

[wmf → notatoad]

Did you read the article? TCP does not deliver line rate 10 Gbps between Japan and Europe, especially with packet loss.

[waps → wmf]

There's a reason it doesn't. Packet loss means the links between Japan and Europe are congesting, and people should back off with their sending speeds.

If they don't, we won't have an internet anymore. This protocol looks designed to make that happen.

[m0nastic]

Superficially, this sounds similar to what CurveCP[1] does (modulo the cryptography being done on the latter).

[1] http://curvecp.org/index.html

[Geee]

The press release is slightly better http://www.fujitsu.com/global/news/pr/archives/month/2013/20...

Basically it's a software solution that transmits over UDP with their own mix of algorithms to make it reliable, but faster than TCP. TCP is broken on mobile, anyway [1]. So, this should be good news.

[1] http://blog.davidsingleton.org/mobiletcp/

[snogglethorpe → Geee]

Er, I'm not so sure it's good news, given their obvious desire to "monetize" it.

A standard protocol that's freely implementable by all is vastly superior to a slightly faster proprietary protocol heavily encumbered by IP bullshit.

[chii → snogglethorpe]

Monetizing a protocol is poor form, but monetizing an implementation of a protocol is not.

If they created a new protocol specification, and opened it up, _and_ their implementation (which could be proprietary for all i care) is faster, then its all fine. As long as other people could implement the protocol in a different way without infringing on patents.

[snogglethorpe → chii]

Sure, but that last "as long as" is a huge one ... getting on a patent gravy-train is almost certainly a prime goal.

[wmf]

Sounds generally similar to Aspera, RaptorQ, etc. It's not clear what's new here, if anything.

[ricardobeat → wmf]

Why aren't we replacing TCP with these?

[wmf → ricardobeat]

Realistically, normal TCP works fine in most cases and multiple parallel TCP connections (a la GridFTP or BitTorrent) can fix the other cases. TCP replacements tend to be pretty complex and some of them are proprietary, so IMO it's just not worth the complexity.

[cicloid → ricardobeat]

Lots of legacy equipment

[politician]

tl;dr: Fujitsu makes an unverifiable claim because the details of the algorithm have not been made public.

[prodigal_erik]

The "bouncer" metaphor seems to be describing http://en.wikipedia.org/wiki/Transmission_Control_Protocol#S..., which we already have. Is it really possible to do that much better?

[insaneirish]

TCP is plenty fast when tuned for the latency of the link. You just need appropriately large window sizes.

[jpxxx]

Oh look, it has better "latancy". Flag and bag.