r1ftrouter on Thu, 30 Jul 2015 19:59:20 +0200 (CEST)


[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: <nettime> Lori Emerson: What's Wrong With the Internet and How


   It is impossible to be able to account for all the requirements around
   delivering a specific product, especially when development is done by
   consensus. Without this, widespread adoption of the technology is going
   to be difficult as people will do any write their own. It is a bit of a
   luxury from a systems perspective that we have a commonality - the OSI
   model by example - from which engineers can agree to build upon. This
   has been a factor in the extent of interoperability for networks in
   general - for better or for worse.

   In terms of aesthetics decisions regarding protocol design - being able
   to do things like error correction, quality-of-service, multiplexing
   multiple sockets over a pair of stacks - are all communications
   technical drivers for building. UDP offers a minimalist alternative but
   you rely on the application developer to code error correction on their
   own - an onerous task hence the various messaging libraries available
   for use. We can reflect back on things like Appletalk, IPX/SPX, Banyan
   VINES and critically question why these stacks lost out over TCP/UDP.
   Current questions engineers ask themselves are how to multipath tcp
   connections while preserving the integrity of the data at the receiver.

   The reordering on the destination stack depending on stream bit rate
   becomes compute intensive and lossy at high packet rates. Protocols
   that are pure signaling based that _make the assumption that the
   receiver has all the data_ by the use of more sophisticated AI methods
   could drastically reduce the amount of network utilization. The most
   elegant protocols have no overhead, wasted bits, and completely deaf
   unless there's something to do.

   With the direction of compute it seems like networks will adopt
   something like an RDMA variant with endpoints just addressing direct
   memory (obviously challenging since a programmers job of tracking
   memory addresses is very hard and wants the OS to do it for them).

   //r1ftrouter
   //network images
   //@r1ft__rout3r


#  distributed via <nettime>: no commercial use without permission
#  <nettime>  is a moderated mailing list for net criticism,
#  collaborative text filtering and cultural politics of the nets
#  more info: http://mx.kein.org/mailman/listinfo/nettime-l
#  archive: http://www.nettime.org contact: nettime@kein.org