The future of the Internet as of 1995

Scott Shenker, "Fundamental Design Issues for the Future Internet"

I was amused by the beginning of the article: 3000000 hosts and astonishment that a newspaper has an online column...

The goal of the paper is to build a framework to analyze future techniques deployed on the Internet to solve problems arising from "real-time" applications.

Current (as of 1995):

• best-effort, no admissions control, no assurance about delay or delivery
• killer apps: traditional data uses (telnet, FTP, HTTP, DNS, SMTP)
• apps are elastic, i.e. tolerant to delay or loss, gracefully degrade in these cases
• apps can change transmission rates in case of congestion

Future:

• "Real-time" apps: video & audio
• apps not elastic, bad performance in case of variation of delays (jitter) or loss
• apps without congestion control
• "Realt-time" apps interfere badly with traditional data uses, no fairness
  

Possible Solutions:

• changing the router implementation, e.g. FQ
• changing application implementation, e.g. delay adaptive techniques
• introduce "Qualitiy of Service": implicit (router analyzes traffic patterns) or explicit (in header)
• Admissions Control
• Overprovisioning

Metrics:

• service delivered to application i encoded in vector s_i
• U_i(s_i) gives performance of application, e.g. U_i for FTP would depend smoothly and mostly linearly on bandwidth, but for a video conference it would degrade when there is not enough bandwidth
  
• total performance V=sum U_i(s_i) is to be maximized
• V does not account for fairness!
  

The possible solutions are discussed in view of the used metrics in "Gedankenexperimenten".