On the Robustness of Soft State Protocols

On the Robustness of Soft State ProtocolsB00902028B00902071Reading materials On the Robustness of Soft State ProtocolsAuthor: John C.S. Lui, Vishal Misra, Dan Rubenstein Outline IntroductionMotivationThe impact of the refresh timer valueCorrelated, lossy feedback channelBroadcast flooding ConclusionCommunication protocolA procedure used by 2 interacting parties to exchange information

Often maintain a modifiable state that is used to track the progress of communicationsoftness of a protocol stateDefault value : A communication endpoint's state can take on several valuesReverts to its default value if the endpoint receives no communication within a time period.Commonly referred to as the timeout period.

softness of a protocol stateThe manner in which the timeout mechanism is usedHard state: used only as an emergency failsafeSoft state: utilize timeout mechanism under normal operating conditionThe principle differencethe time over which the refresh timer expiresthe soft state protocol requires a small amount of additional time after the session officially endsfor the refresh timer to expire and reset the sender to its default state.robustnessThe protocol's performance under a variety of network conditions is above an acceptable threshold.

Need not be optimalrobustness

characteristics of soft state protocolSoft state protocols are less trusting of misbehaving endpoints.

DOS attack

characteristics of soft state protocolSoft state protocols use the timeout mechanism to create a virtual, predictable, feedback channel.

Correlated lossy feedback channel

characteristics of soft state protocolSoft state protocols are less likely to flood the network with signaling traffic

Broadcast floodingThe impact of refresh timer valueApplication-specific inconsistency cost.State (Re)Initialization cost.Refresh overhead.Stale state cost.

Analytical modelLife time: LI:initialization stateV: valid stateTD: tear downS: stale stateRefresh every R time unit

Choosing the right RThe principle difference between soft and hard state protocolRefresh and teardown message has a loss probability pNumber of refresh event: L/RMeasuring performance by calculating cost C(R)

Calculating Cost#loss refresh message: pL/RProtocol enters an inconsistent stateTradeoff between long time spent in inconsistent state ,or more refresh messages

Calculating CostA reinitialize cost when entering inconsistent stateTime spent in inconsistent state: pL

#(re)initialization : 1+pL/R

Calculating CostStale state cost:Proportion to refresh time interval RFor soft state protocol ,this cost is incurred every time a state expiresFor hard state protocol, it is incurred when a state removal message gets lostCaculating Cost

Calculating Cost

Fundamental hard vs soft tradoffComparisonX-axis: loss probabilityY: cost

DOS attackDenial of serviceAn attacker reserves a resource , and then exiting without tearing down the connectionLife time = 0 ,channel loss probabilty = 1Comparison

A Correlated, Lossy Feedback ChannelProblemcongestion collapseModel Initial new sessions with rate Mean time complete the transfer of data is 1/ active session and inactive session

Sever can be simultaneously host up to N sessionsChannel can support N < M sessions without loss A Correlated, Lossy Feedback Channel (cont.) Hard state protocol Client attempts n times to deliver a message to the server to terminate the connection soft state fail safe: a time with mean 1/t after the client aborts, server can sense the absence of client and terminateSoft state protocol clients ping the server at high rate when the server does not receive the ping for a time with mean 1/s , it terminates1/s