On Thu, Sep 11, 2014 at 9:38 AM, Matthew Lohbihler <[email protected]> wrote: > Hi Chris, > > Not sure what you mean. Any data that can be digitized can be sent over IP - > no issue here. All i meant to suggest was that to improve the routing of > this data over the internet though would, i believe, require more > information than a router currently has access to.
No routers use a routing protocols between themselves to tell each other what other routers they have access to. The DO have this information. They can compute shortest paths and best paths and will quickly see when a router goes down see this for more info: http://www.cisco.com/c/en/us/td/docs/net_mgmt/active_network_abstraction/3-7/reference/guide/ANARefGuide37/routpro.html > For example, knowing that > a particular node is down or a typical route is congested is something that > sufficiently distant routers would not know, and therefore they would not be > able to change their behaviour accordingly. Also, i don't think IP has any > notion of application-level prioritization (i.e. video over email), which > was mentioned in the article. Yes IP does have the concept of priority here is a link to the relevant standard. It is very widely implemented and used for the exact purposes you describe http://www.ietf.org/rfc/rfc2475.txt > > Regards, > Matthew > > > On 9/11/2014 11:46 AM, Chris Albertson wrote: > > Can you say exactly what kind of data currently can not be sent over IP? > > Remember that IP is not limited to only TCP there is also UDP and > others. IP can also handle broadcasts. It is not limited to point to > point. > > What a change can never do is improve latency and speed. That > requires improved hardware. So aside from speed what exactly can't > the current IP stack do? > > On Thu, Sep 11, 2014 at 7:01 AM, Matthew Lohbihler > <[email protected]> wrote: > > I think this would be a very interesting problem to work on. But it seems to > me to implement what the article is talking about would require changes to > the internet protocol, which of course would be a long and tortuous process. > And without knowing what changes to IP will be approved, we can't know what > data NuPIC would have available for analysis. > > It's a chicken and egg problem. You have to iterate through a process of IP > and routing software changes in order to test improvement proposals. And at > the end of your iterations, you have both the necessary IP changes as well > as the routing software. NuPIC could be used within the iterations i > suppose, but there's still a lot of old-fashioned hard work to do. > > > On 9/8/2014 1:24 PM, Matthew Taylor wrote: > > Rik has a great point here (and a nice article). If someone can figure out > how to use NuPIC to solve this problem, we'd be the darlings of the internet > (for a short period of time). > > Is anyone else thinking along these lines? > > --------- > Matt Taylor > OS Community Flag-Bearer > Numenta > > On Sun, Sep 7, 2014 at 10:34 PM, Rik <[email protected]> wrote: > > Everyone > > If you're not thinking hard about how to put machine intelligence to work > in computer networking, then you're doing the wrong thing. Leave your robots > and other anthropocentric toy projects aside and work on stuff that matters! > > Some good pep talk: > http://spectrum.ieee.org/computing/networks/why-the-internet-needs-cognitive-protocols > > A good example how the 'motor' in 'sensory-motor integration' can and > should be taken to be more metaphorical rather than an actual motor. The > 'action' an intelligent agent on the network takes is to send some data down > a pipe. This may eventually result in a motor action at the other end of the > pipe but that's taken care of, plenty of people are busy hooking up motors > to the 'net already. Who does the cognitive bit? > > Cheers > > Rik > > > > > > -- Chris Albertson Redondo Beach, California
