Survivability Quantification of Communication Services Poul E.Heegaard, Kishor S. Trivedi Advisor: Frank Y. S. Lin Presented by Y.W. Lee.

Slides:



Advertisements
Similar presentations
Algorithm Analysis Input size Time I1 T1 I2 T2 …
Advertisements

Discrete time Markov Chain
Chapter 13 Queueing Models
Based on: Petri Nets and Industrial Applications: A Tutorial
Queuing Network Models for Delay Analysis of Multihop Wireless Ad Hoc Networks Nabhendra Bisnik and Alhussein Abouzeid Rensselaer Polytechnic Institute.
Lecture 6  Calculating P n – how do we raise a matrix to the n th power?  Ergodicity in Markov Chains.  When does a chain have equilibrium probabilities?
Queueing Models and Ergodicity. 2 Purpose Simulation is often used in the analysis of queueing models. A simple but typical queueing model: Queueing models.
1 A class of Generalized Stochastic Petri Nets for the performance Evaluation of Mulitprocessor Systems By M. Almone, G. Conte Presented by Yinglei Song.
IE 469 Manufacturing Systems
Delay and Throughput in Random Access Wireless Mesh Networks Nabhendra Bisnik, Alhussein Abouzeid ECSE Department Rensselaer Polytechnic Institute (RPI)
Topics Review of DTMC Classification of states Economic analysis
1 Part III Markov Chains & Queueing Systems 10.Discrete-Time Markov Chains 11.Stationary Distributions & Limiting Probabilities 12.State Classification.
Markov Reward Models By H. Momeni Supervisor: Dr. Abdollahi Azgomi.
Lecture 13 – Continuous-Time Markov Chains
Models and Security Requirements for IDS. Overview The system and attack model Security requirements for IDS –Sensitivity –Detection Analysis methodology.
Nur Aini Masruroh Queuing Theory. Outlines IntroductionBirth-death processSingle server modelMulti server model.
Synthesis of Embedded Software Using Free-Choice Petri Nets.
A. BobbioBertinoro, March 10-14, Dependability Theory and Methods 5. Markov Models Andrea Bobbio Dipartimento di Informatica Università del Piemonte.
1 Performance Evaluation of Computer Networks Objectives  Introduction to Queuing Theory  Little’s Theorem  Standard Notation of Queuing Systems  Poisson.
Performance Analysis of Wavelength-Routed Optical Networks with Connection Request Retrials Fei Xue+, S. J. Ben Yoo+, Hiroyuki Yokoyama*, and Yukio Horiuchi*
A General approach to MPLS Path Protection using Segments Ashish Gupta Ashish Gupta.
Modeling and Analysis of Manufacturing Systems Session 2 QUEUEING MODELS January 2001.
7/3/2015© 2007 Raymond P. Jefferis III1 Queuing Systems.
Queuing Networks: Burke’s Theorem, Kleinrock’s Approximation, and Jackson’s Theorem Wade Trappe.
A General approach to MPLS Path Protection using Segments Ashish Gupta Ashish Gupta.
1 A State Feedback Control Approach to Stabilizing Queues for ECN- Enabled TCP Connections Yuan Gao and Jennifer Hou IEEE INFOCOM 2003, San Francisco,
Petri Nets An Overview IE 680 Presentation April 30, 2007 Renata Kopach- Konrad.
Buffer Management for Shared- Memory ATM Switches Written By: Mutlu Apraci John A.Copelan Georgia Institute of Technology Presented By: Yan Huang.
1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.
Introduction to Discrete Event Simulation Customer population Service system Served customers Waiting line Priority rule Service facilities Figure C.1.
1 Performance Evaluation of Computer Networks: Part II Objectives r Simulation Modeling r Classification of Simulation Modeling r Discrete-Event Simulation.
Adviser: Frank, Yeong-Sung Lin Present by Wayne Hsiao.
Capacity analysis of complex materials handling systems.
Generalized Semi-Markov Processes (GSMP)
Chapter 2 Machine Interference Model Long Run Analysis Deterministic Model Markov Model.
Topology aggregation and Multi-constraint QoS routing Presented by Almas Ansari.
Department of Information Engineering University of Padova, ITALY Performance Analysis of Limited–1 Polling in a Bluetooth Piconet A note on the use of.
Lecture 14 – Queuing Networks Topics Description of Jackson networks Equations for computing internal arrival rates Examples: computation center, job shop.
Decision Making in Robots and Autonomous Agents Decision Making in Robots and Autonomous Agents The Markov Decision Process (MDP) model Subramanian Ramamoorthy.
Network Survivability Against Region Failure Signal Processing, Communications and Computing (ICSPCC), 2011 IEEE International Conference on Ran Li, Xiaoliang.
Queuing Theory Basic properties, Markovian models, Networks of queues, General service time distributions, Finite source models, Multiserver queues Chapter.
1 Elements of Queuing Theory The queuing model –Core components; –Notation; –Parameters and performance measures –Characteristics; Markov Process –Discrete-time.
1 Chapters 8 Overview of Queuing Analysis. Chapter 8 Overview of Queuing Analysis 2 Projected vs. Actual Response Time.
Why Wait?!? Bryan Gorney Joe Walker Dave Mertz Josh Staidl Matt Boche.
Generalized Semi- Markov Processes (GSMP). Summary Some Definitions The Poisson Process Properties of the Poisson Process  Interarrival times  Memoryless.
Generalized stochastic Petri nets (GSPN)
Modelling by Petri nets
Chapter 20 Queuing Theory to accompany Operations Research: Applications and Algorithms 4th edition by Wayne L. Winston Copyright (c) 2004 Brooks/Cole,
CS433 Modeling and Simulation Lecture 07 – Part 01 Continuous Markov Chains Dr. Anis Koubâa 14 Dec 2008 Al-Imam.
School of Computer Science, The University of Adelaide© The University of Adelaide, Control Data Flow Graphs An experiment using Design/CPN Sue Tyerman.
OPERATING SYSTEMS CS 3530 Summer 2014 Systems and Models Chapter 03.
CAP 4800/CAP 5805: Computer Simulation Concepts
(C) J. M. Garrido1 Objects in a Simulation Model There are several objects in a simulation model The activate objects are instances of the classes that.
CS433 Modeling and Simulation Lecture 11 Continuous Markov Chains Dr. Anis Koubâa 01 May 2009 Al-Imam Mohammad Ibn Saud University.
11. Markov Chains (MCs) 2 Courtesy of J. Bard, L. Page, and J. Heyl.
8/14/04J. Bard and J. W. Barnes Operations Research Models and Methods Copyright All rights reserved Lecture 11 – Stochastic Processes Topics Definitions.
Petri-Nets and Other Models
Research Direction Introduction Advisor: Frank, Yeong-Sung Lin Presented by Hui-Yu, Chung 2011/11/22.
Distance Vector Routing
Simulation Examples And General Principles Part 2
State Modeling. Introduction A state model describes the sequences of operations that occur in response to external stimuli. As opposed to what the operations.
OPERATING SYSTEMS CS 3502 Fall 2017
Lecture 14 – Queuing Networks
Availability Availability - A(t)
Dynamic Graph Partitioning Algorithm
Hidden Markov Models Part 2: Algorithms
IV-2 Manufacturing Systems modeling
Lecture 14 – Queuing Networks
Buffer Management for Shared-Memory ATM Switches
Presentation transcript:

Survivability Quantification of Communication Services Poul E.Heegaard, Kishor S. Trivedi Advisor: Frank Y. S. Lin Presented by Y.W. Lee

Agenda Abstract Petri Net and Stochastic Reward Net Network of waiting lines Phase dependent performance Space decomposed model Network example Conclusion

Agenda Abstract Petri Net and Stochastic Reward Net Network of waiting lines Phase dependent performance Space decomposed model Network example Conclusion

Abstract Petri net vs Stochastic Reward Net Exchange information Independent in waiting lines Decomposition and performance

Abstract Objective: quantify the survivability of virtual connections in telecommunication. – service: the VC between specific peering nodes in the network. – requirement: maximum packet loss probability and end-to-end delay of non-lost packets in the VC. – undesired events: link and node failures caused by attacks.

Agenda Abstract Petri Net and Stochastic Reward Net Network of waiting lines Phase dependent performance Space decomposed model Network example Conclusion

Petri net A Petri net is one of several mathematical modeling language for the description of discrete distributed systems. A Petri net consists of places, transitions, and directed arcs.

Places : – Contain any non-negative number of tokens. – Input place before transition and output place after transition. Marking : – A distribution of tokens over the places of a net. Directed arcs : – Run between places and transitions. – Each arc has multiplicity. (bandwidth) Transitions : – Fire whenever there is a token at the end of all input arcs. Fire : – It consumes these tokens, and places tokens at the end of all output arcs. – Duration : 0.

A Petri net graph is a 4-tuple (S,T,W, ), where – S is a finite set of places – T is a finite set of transitions – S and T are disjoint – W : define an arc and each arc has an non-negative integer arc multiplicity – The preset of a transition t is the set of its input places: – Its postset is the set of its output places: – : initial marking

Its transition relation can be described as a pair of | S | by | T | matrices: – W −, defined by – W +, defined by Then their difference W T = W + − W − can be used to describe the reachable markings in terms of matrix multiplication.

For any sequence of transitions w, write o(w) for the vector that maps every transition to its number of occurrences in w. Then, we have } is a firing sequence of N

- = =

SRN The SRN differ from SPN (PN with stochastic transition rate) in several key aspects : – Enable function (in transition) – Marking-dependent arc cardinalities – The ability to decide in a marking-dependent fashion whether the firing time of a transition is exponentially distributed or null

We define a non-parametric SRN as an 11- tuple : the inhibitor arc from p to t. when the transition t fires in marking μ,the new marking satisfies:

 {true, false} is the guard associated to transition t. > is a transition and irreflexive imposing a priority among transitions. is the initial marking. is the rate of the exponentially distribution for the firing time of transition t.

The definition of vanishing and tangible marking – Vanish : at least one transition time duration is immediate. – Tangible : all transition time duration is timed.

We assume a race model marking μ occupied by transition t in probability The firing probability of enabled transition t is: : In a tangible marking μ : In a vanish marking μ The notation indicates that transition t is enabled in marking μ.

The elements described so far define a trivariate discrete-time stochastic process: (,, ) – : the n-th transition to fire. – : the time at which n-th fire. – : the n-th marking encountered.

is a finite set of measure. – ρ( μ ) reward rate : the rate at which reward is accumulated when the marking is μ. – reward impulse : the instantaneous reward gained when firing transition t while in marking μ. – : compute a single value from Y(θ) as index.

Reward process accumulated by the SRN to time θ Number of transition fires up to time θ :E(Y(θ)) – Expect time average reward up to time θ In steady state:

Exchange of data Given a SRN A – Define c as a boolean marking-dependent expression (1: hold 0: not hold), : tangible marking according to whether the condition c is on or off. : steady-state probability.

Assume that each node is an M/M/1/-queue. is the steady state probability that node i will reject an incoming packet. (2)(2)

: transition rate at which condition goes from on to off. : sum of steady state probability of each marking in.

: the expected time to absorption when the marking in are considered absorbing, starting from steady state. : the expected time spent in each non-absorbing tangible marking before absorbing., : and of the states in.

= [0.1,0.2,0.3,0.4] ={1,2}, ={3,4}

If condition b holds in the set of marking we are interested in, we can use a modified version of the steady state vector, define as:

The quantities, and can be generalized to, and.

The following steps are needed to exploit near-independence at the SRN level – Decomposition – Import graph Fix the value of when we want to compute,if > – Iteration Acyclic Cycle

Agenda Abstract Petri Net and Stochastic Reward Net Network of waiting lines Phase dependent performance Space decomposed model Network example Conclusion

Network of waiting lines A machine shop has several departments each containing a fixed number of identical machines. Each department is a multiserver system of the usual type. Arrivals at a given department come both form other departments in the shop and from outside the shop.

If mean arrival rates at the various departments are properly defined. Result: steady-state distribution in which the waiting-line lengths of the departments are independent.

Single department : k : customer number Serve with n identical servers

M department ……? – Department m contain servers. – Customs arrive from outside with Poisson-type – FCFS and service time: – Once served in department m, a customer goes instantaneously to department k with P( ) ; his total service is completed with probability :

: the average arrival rate of customers at department m from any source. A steady-state distribution of the state of the system is given by the products

Agenda Abstract Petri Net and Stochastic Reward Net Network of waiting lines Phase dependent performance Space decomposed model Network example Conclusion

The phased recovery model describes the cycle from failure until the system is back to steady state. Each phase may have different set of available resources for the virtual connections. Represented as phase-dependent stationary routing probabilities with corresponding phase-dependent arrival rates.

Phased recovery model Phase IV : After the routing information is restored the network operates in fault free mode. – Absorbing state for the purpose of survivability analysis. Phase I : Immediately after the failure the procedure is activated but it takes some time before the rerouting is effective. – packets are routed according to the original routing scheme. –, except for the failed node i and link [i, j] where q ij (I) = 0. – The rerouting time is exponentially distributed with rate.

Phased recovery model Phase II : When the rerouting is effective the link or node is still failed. – new routing scheme avoid these failed links or nodes. – the exponentially distributed repair time with rate. Phase III : On completion of repair the system returns to failure free state but the routing is yet to change. – Phase II and III may have identical routing probabilities with. – exponentially distributed rerouting time with rate.

Performance metrics In this paper the performance metric M includes – The transient loss rate L(t) at time t. – The loss probability l(t) at time t. – The number of packets in the system N(t) at time t. – The mean end-to-end delay D(t) of packets that are not lost in the virtual connection.

Performance metrics

The transient probabilities,, are obtained from the composite models. : The external arrival rate.

Performance metrics

Agenda Abstract Petri Net and Stochastic Reward Net Network of waiting lines Phase dependent performance Space decomposed model Network example Conclusion

Exact network survivability models The modeling assumptions in the simulation and Stochastic Reward Net (SRN) models are identical with – exponentially distributed interevent times – time-independent – phase-dependent routing probabilities.

Stochastic Reward Net model The packets are tokens that are generated by the timed transition “arrival” into the place “InQ1”. If there are less than tokens in place “Node1” the immediate transition “Q1” is enabled. If not, the “loss1” transition is enabled and count the loss packet number. The routing is determined by probabilities on the immediate transitions out “OutQ1”.

The rerouting, failure and repair are modeled at the top of the SRN model where the tokens in the “phase y” places will constrain the token passing of the failed node through inhibitor arcs as illustrated in the figure.

We decompose the problem and approximate the global probabilities by the product.

Calculate the arrival rates to each node in the queuing network for VC v in node i by solving the linear system of traffic equations:

Assume that each node is an M/M/1/-queue. is the steady state probability that node i will reject an incoming packet. (2)(2)

Space decomposed model As per our space decomposition approximation we model the transient behavior in each node separately. To give an example consider the four node case in right where node j = 2 has failed and the non-failed nodes (i = 1, 3, 4).

all the packets that are sent to node j are lost, hence all transitions lead to state ( ) where all resources are unavailable and no packets will be served.

Immediately after the undesired event the network state is changed from to. no packets are lost but the arrival rates are changed to by (1)

Space decomposed model In the model of the failed node j the transient probability is obtained with the initial condition For the non-failed nodes (i ≠ j) the is obtained with the initial condition

Finally, the global state probabilities are obtained by product form approximation (3)(3)

Agenda Abstract Petri Net and Stochastic Reward Net Network of waiting lines Phase dependent performance Space decomposed model Network example Conclusion

A 4 node example The first example is a network with n = 4 nodes. The performance of the virtual connection between s = 1 and d = 4 is evaluated after the failure of node 2 at time t = 500. Each node i is an M/M/1/ni system with the parameters given in Table 1. The parameters in the phased recovery model are and.

(a) Loss probability, l(t)(b) Mean number, N(t) The estimated performance metrics from R = 90 simulation replica (Simulations) are compared against the analytical values of the Stochastic Reward Net model solved by SPNP (SRN model).

A 10 node example The directed graph G[1,10] for routing virtual connections between s = 1 and d = 10. The performance of the virtual connection is evaluated after the failure of node 4 at time t = 500. each node is an M/M/1/ni system with the parameters given in Table 2. The parameters in the phased recovery model are and.

(a) Loss probability, l(t) (b) Number in the system, N(t) In this example also includes a “rerouting model” which is, i.e. the probability that a packet is lost in the failed node at time t after the instant of failure.

A 10 node example The results in Figure 9(b) means with very low steady-state packet loss probability, the approximation is good. The transient loss probability is dominated by q(1, 4) with the decay rate equal to the reciprocal of the expected rerouting time. The same is not observed in the 4-node network because here steady-state loss probability is not negligible.

Agenda Abstract Petri Net and Stochastic Reward Net Network of waiting lines Phase dependent performance Space decomposed model Network example Conclusion

In our space decomposition model we assumed that : – Independence between the network nodes which is not a fully realistic assumption but a good approximation in networks with low loss probability and with high aggregation. – In each phase we have steady-state performance : Performance in a phase is reached quickly after the change of phase compared to the expected duration of the phase.

The approximation is good if there is at least two order of magnitude difference between the time granularity of the events in the performance model and in the recovery model.

in medium loaded (30-50%) high capacity networks (100Mbit/s -10 Gbit/s) we will observe packets/ms, while the routing, rerouting and repair (at IP level) is in the order of 100s of ms.  This means a few hundred to several thousand packets are expected in each phase.

The results show that when the transient performance is dominated by impairments in a single node this decomposed, product form approximation is a viable approach.

The assumptions made in our models will be relaxed in the future allowing for multiple failures, general distribution and multiple virtual connections. We plan to extend the phased recovery model to include more details regarding the virtual connection management at failure and repair, and possibly to include multiple failure modes.

Thanks for your listening !