This paper presents a fair queueing algorithm designed to give each network conversation a fair share. When used with other endhost based congestion avoidance scheme, fair queueing can protect the network from malicious ill-behaved users. In addition, this scheme allows independent allocation policies for promptness, buffer and bandwidth . This leads to a faster response time for interactive sessions such as Telnet sessions.
To implement fair queueing algorithm at the gateway level, gateways are forced to keep flow level information. This was less than ideal because these state manipulation adds latency to the link, and makes fault recovery difficult to implement, which is the exact motivation of the next paper. However, in today's world, the need to keep flow level state information is growing because of traffic engineering and security needs. The balance between functionality vs. latency and reliability is likely going to depend on the operating environment and the particular application needs.
The definition of conversation is contriversial till this day. Could there be a better way to associate traffic with a particular user? What are the privacy implications of such association? I am curious to find out how many of current abuse of networks stems from the inability of network gateways to define user exactly.
This paper lays good foundation of what fair queueing is and serves as good background material for the next paper.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment