Tuesday, October 16, 2012

In Praise of Deterministic "Flows"

Many of you are too young to remember the fierce debates about the principles that "should" guide the development of the "next generation network." That was before Internet Protocol stopped being an "interim" or "legacy" protocol and seemingly become the foundation for all next generation networks.

Telco architects had a different vision, based on the asynchronous transfer mode protocol, and popularly known at times as "broadband ISDN." Integrated services digital network was an earlier generation of thinking about the "next generation network." 

"Open Systems Interconnect" was supposed to replace IP. It didn't, obviously. But the backers of ATM knew something important, namely that some applications and services can tolerate delay and latency and packets arriving out of order. 

But some applications and services really do not work very well unless packets arrive in order, and that can be very challenging when so many packets have to arrive, so rapidly, that simple error correction methods, such as re-transmitting, do not work. 

That is true for video, and especially for high-definition video. Other services, such as real-time voice, likewise are intolerant of packets arriving out of order, because it is not elegant to retransmit the garbled bits and reinsert them into the stream fast enough to preserve quality.

In other words, the architects behind ATM knew something important about at least some forms of network-transmitted content, sessions and information, and that was that "best effort" often does not work for isochronous traffic. 

In other words, engineers now have to work around the "best effort" architecture of IP to replicate what would be built in to ATM, namely quality of service measures that handle different flows of information in a differential manner. 

Some will argue IP can be made to work that way. Others see problems ahead. 

 

No comments:

Which Firm Will Use AI to Boost Revenue by an Order of Magnitude?

Ultimately, there is really only one way for huge AI infrastructure investments up by an order of magnitude over cloud computing investment ...