Long Range Dependent Traffic and Leaky Buckets CS215-winter ’01 Demetrios Laios 3/22/2001.

Slides:



Advertisements
Similar presentations
CprE 458/558: Real-Time Systems
Advertisements

Martin Suchara, Ryan Witt, Bartek Wydrowski California Institute of Technology Pasadena, U.S.A. TCP MaxNet Implementation and Experiments on the WAN in.
Internet Measurement Conference 2003 Source-Level IP Packet Bursts: Causes and Effects Hao Jiang Constantinos Dovrolis (hjiang,
Connection-level Analysis and Modeling of Network Traffic understanding the cause of bursts control and improve performance detect changes of network state.
1 CONGESTION CONTROL. 2 Congestion Control When one part of the subnet (e.g. one or more routers in an area) becomes overloaded, congestion results. Because.
CONGESTION CONTROL T.Najah Al-Subaie Kingdom of Saudi Arabia Prince Norah bint Abdul Rahman University College of Computer Since and Information System.
Transport Layer3-1 TCP AIMD multiplicative decrease: cut CongWin in half after loss event additive increase: increase CongWin by 1 MSS every RTT in the.
24-1 Chapter 24. Congestion Control and Quality of Service (part 1) 23.1 Data Traffic 23.2 Congestion 23.3 Congestion Control 23.4 Two Examples.
CS640: Introduction to Computer Networks Aditya Akella Lecture 20 – QoS.
Traffic Shaping Why traffic shaping? Isochronous shaping
CS Spring 2009 CS 414 – Multimedia Systems Design Lecture 16 – Multimedia Transport Subsystem (Part 3) Klara Nahrstedt Spring 2009.
ATM : CONGESTION CONTROL Group 7 29 DECEMBER 2004.
1.  Congestion Control Congestion Control  Factors that Cause Congestion Factors that Cause Congestion  Congestion Control vs Flow Control Congestion.
Jaringan Komputer Lanjut Traffic Management Aurelio Rahmadian.
Priority Scheduling and Buffer Management for ATM Traffic Shaping Authors: Todd Lizambri, Fernando Duran and Shukri Wakid Present: Hongming Wu.
NETWORK LAYER. CONGESTION CONTROL In congestion control we try to avoid traffic congestion. Traffic Descriptor Traffic descriptors are qualitative values.
Recent Progress on a Statistical Network Calculus Jorg Liebeherr Department of Computer Science University of Virginia.
Comments on the Performance of Measurement Based Admission Control Algorithms Lee Breslau, S. Jamin, S. Shenker Infocom 2000.
Comparison and Analysis of FIFO, PQ, and WFQ Disciplines on multimedia
Advanced Computer Networking Congestion Control for High Bandwidth-Delay Product Environments (XCP Algorithm) 1.
CMPT 855Module Network Traffic Self-Similarity Carey Williamson Department of Computer Science University of Saskatchewan.
Dynamic Internet Congestion with Bursts Stefan Schmid Roger Wattenhofer Distributed Computing Group, ETH Zurich 13th International Conference On High Performance.
On the Self-Similar Nature of Ethernet Traffic - Leland, et. Al Presented by Sumitra Ganesh.
End-to-End Analysis of Distributed Video-on-Demand Systems Padmavathi Mundur, Robert Simon, and Arun K. Sood IEEE Transactions on Multimedia, February.
1 Random Early Detection Gateways for Congestion Avoidance Sally Floyd and Van Jacobson, IEEE Transactions on Networking, Vol.1, No. 4, (Aug 1993), pp
Self-Similarity in Network Traffic Kevin Henkener 5/29/2002.
Random Early Detection Gateways for Congestion Avoidance
Advanced Computer Networks : RED 1 Random Early Detection Gateways for Congestion Avoidance Sally Floyd and Van Jacobson, IEEE Transactions on Networking,
Self-Similarity of Network Traffic Presented by Wei Lu Supervised by Niclas Meier 05/
1 Chapters 9 Self-SimilarTraffic. Chapter 9 – Self-Similar Traffic 2 Introduction- Motivation Validity of the queuing models we have studied depends on.
References for M/G/1 Input Process
A Generalized Processor Sharing Approach to Flow Control in Integrated Services Networks: The Single-Node Case Abhay K. Parekh, Member, IEEE, and Robert.
Dynamic versus Static Traffic Policing: A New Approach for Videoconference Traffic over Wireless Cellular Networks Author: Polychronis Koutsakis Anukrati.
QOS مظفر بگ محمدی دانشگاه ایلام. 2 Why a New Service Model? Best effort clearly insufficient –Some applications need more assurances from the network.
“A non parametric estimate of performance in queueing models with long-range correlation, with applications to telecommunication” Pier Luigi Conti, Università.
1 Why traffic shaping? yIn packet networks that implement resource sharing xadmission control and scheduling alone are insufficient users may attempt to.
MIT Fun queues for MIT The importance of queues When do queues appear? –Systems in which some serving entities provide some service in a shared.
1 Optical Burst Switching (OBS). 2 Optical Internet IP runs over an all-optical WDM layer –OXCs interconnected by fiber links –IP routers attached to.
Network Instruments VoIP Analysis. VoIP Basics  What is VoIP?  Packetized voice traffic sent over an IP network  Competes with other traffic on the.
ACN: RED paper1 Random Early Detection Gateways for Congestion Avoidance Sally Floyd and Van Jacobson, IEEE Transactions on Networking, Vol.1, No. 4, (Aug.
1 Lecture 14 High-speed TCP connections Wraparound Keeping the pipeline full Estimating RTT Fairness of TCP congestion control Internet resource allocation.
A test-bed investigation of QoS mechanisms for supporting SLAs in IPv6 Vasilios A. Siris and Georgios Fotiadis University of Crete and FORTH Heraklion,
CS 447 Network & Data Communication QoS (Quality of Service) & DiffServ Introduction Department of Computer Science Southern Illinois University Edwardsville.
1 On Class-based Isolation of UDP, Short-lived and Long-lived TCP Flows by Selma Yilmaz Ibrahim Matta Computer Science Department Boston University.
TCP Trunking: Design, Implementation and Performance H.T. Kung and S. Y. Wang.
9.7 Other Congestion Related Issues Outline Queuing Discipline Avoiding Congestion.
CS640: Introduction to Computer Networks Aditya Akella Lecture 20 - Queuing and Basics of QoS.
Spring 2003CS 3321 Congestion Avoidance. Spring 2003CS 3322 Congestion Avoidance TCP congestion control strategy: –Increase load until congestion occurs,
Burst Metric In packet-based networks Initial Considerations for IPPM burst metric Tuesday, March 21, 2006.
Analysis of RED Goal: impact of RED on loss and delay of bursty (TCP) and less bursty or smooth (UDP) traffic RED eliminates loss bias against bursty traffic.
CONGESTION CONTROL.
Explicit Allocation of Best-Effort Service Goal: Allocate different rates to different users during congestion Can charge different prices to different.
Random Early Detection (RED) Router notifies source before congestion happens - just drop the packet (TCP will timeout and adjust its window) - could make.
CS Spring 2011 CS 414 – Multimedia Systems Design Lecture 16 – Multimedia Transport (Part 2) Klara Nahrstedt Spring 2011.
Providing QoS in IP Networks
Performance Comparison of Ad Hoc Network Routing Protocols Presented by Venkata Suresh Tamminiedi Computer Science Department Georgia State University.
Chapter 10 Congestion Control in Data Networks and Internets 1 Chapter 10 Congestion Control in Data Networks and Internets.
Tel Hai Academic College Department of Computer Science Prof. Reuven Aviv Markov Models for data flow In Computer Networks Resource: Fayez Gebali, Analysis.
QoS & Queuing Theory CS352.
Topics discussed in this section:
Chapter 6 Congestion Avoidance
Buffer Management in a Switch
CONGESTION CONTROL.
Computer Science Division
Congestion Control, Quality of Service, & Internetworking
Javad Ghaderi, Tianxiong Ji and R. Srikant
Chapter-5 Traffic Engineering.
EECS 122: Introduction to Computer Networks Packet Scheduling and QoS
CPSC 641: Network Traffic Self-Similarity
Presentation transcript:

Long Range Dependent Traffic and Leaky Buckets CS215-winter ’01 Demetrios Laios 3/22/2001

Describes the behavior of data source A necessary component of a good queueing-system model is an arrival-process model that reflects behavior of the actual arrival process sufficiently closely to explain and/or predict the effect of the arrival process on the system Informative descriptors of arrival processes The objective of policing policies such as a leaky-bucket scheme is to reduce congestion by restricting the burstiness of the arrival process.

Burstiness is roughly, the tendency of arrivals to occur in clusters, or bursts, separated by periods of relatively few arrivals, has become commonplace. The significance of burstiness is that it has a pronounced adverse effect on queueing-system behavior. For example, a bursty arrival stream may cause the system to undergo excursions into periods of severe congestion. Various descriptors of traffic processes can be found in the literature.

Linear Bounded Arrival Processes (LBAP) The use of bounded burstiness inputs has been introduced by Cruz (1987) AN LBAP-constrained source bounds the number of bits it transmits in any interval of length  by a linear function of t. Number of bits transmitted in any interval of length    + b –  is roughly the long-term average rate allocated by the net to the source and –b is the maximum burst may sent given the choice of  Leaky bucket scheme regulates a LBAP source Count is incremented by 1 every 1/  seconds, up to a maximum of b

LBAP Cont. Leaky bucket analogy –packets waiting for a permit one permit added every 1/  seconds –packets with permits awaiting transmission holds only b permits The bucket size b is an important parameter in determining the effectiveness of the leaky bucket scheme Small b delays bursty traffic Large b can lead to downstream congestion and buffer overflow

The Leaky Bucket Scheme token bucket size ii transmission tokens connection i bibi output data buffer leaky bucket enforces A i (t 1,t 2 )  b i +  i (t 2 -t 1 )

Properties of Leaky Bucket can thing of this as a single-server queueing system with constant service rate can act as -peak rate if  = peak rate, b = 1 -or moving-window average rate regulator if  = avg rate, b = 1 can act as policer (no packet buffer) Allows to sent a maximun number of packets given the bucket at the max bit rate Performance (loss rate) depends only on the sum of the packet buffer size and the token bucket size (one trades off other exactly)

Traffic bounds The model bounds the traffic (rather than exactly characterize) Several different bounds can be obtained for any traffic source (in the same model): -trade off: token bucket size (b value) against bandwidth share (  value) More bucket depth: larger (worst-case) delays; more jitter More bandwidth: lower overall utilization; future connections more likely to be rejected

Choosing LBAP parameters If b slightly < K 1   increase If  slightly < K 2  b is larger A, average rate P, peak rate  b Knee point K2K2 K1K1 P A  ( , b)=0

Long-memory (self similar) model Studies show that packet traffic is strongly auto-correlated and there exists a long-range dependency (LRD), i.e. persistence in their correlation structured does not die even for large lags. Suppose X, second order (weak or covariance or wide-sense) stationary stochastic process –mean  = E[X t ] –variance E[(X t -  ) 2 ] =  0 –autocovariance  k = E[(X t -  ) (X t+k -  )], k = 0, 1,..,  k =  k /  0 Short-range dependence, or process with short memory or short range correlations or weak dependence –SRD:  k  k <  Long-range dependence, or process with long memory or long range correlations or strong dependence –LRD:  k  k   i.e. diverges

Shuffling the time series of original traffic trace Transform the original traffic trace, preserving the marginal distribution, peak, average rate while destroying correlations External shuffle: –Divide the original sequence into blocks of size M. With n data times, there are n/M such blocks. –Then,while preserving the sequence inside each block, the order of blocks is shuffled –Effect: destroying the long-range correlations in data, while preserving the short-range correlations Internal shuffle: –Now, while preserving the sequence of blocks, a sequence of interarrival times inside each block, is shuffled. –Effect: destroying the short-range correlations in data, while preserving the long-range correlations Total shuffling: destroy short and long range correlation

Statistical bound P[A(t 1,t 2 )   (t 2 - t 1 ) + b] <= const where A(t 1,t 2 ) is written in terms of Fractional Brownian motion A(t,  ) = m  + B H (t +  ) - B H (t)= d m  + B H (  ),  = t 2 - t 1 H : Hurst parameter m : average source bit rate Applying Large Deviations Techniques we get the following form of LBAP curve  ( ,b)=0,  ( ,b) = A( , H)(  -m) 2H -  b 2-2H where, A( , H) = [2  2 H 2H (1-H) 2-2H ] -1

Network Power A measure of the efficiency of the congestion control scheme –Power = (Throughput) a /delay Chose the exponent a based on the relative emphasis placed on throughput versus delay: –if throughput is more important, then a value of a greater than one is chosen. –if throughput and delay are equally important, then a equal to one is chosen. Clearly, we wish to have lots of throughput and small delay. Unfortunately, these cannot both be achieved simultaneously, and so we are looking for the “operating point” for the system.

Design problem (work in progress) Maximize power f ( ,b) =  2 /b Such that: –LBAP curve :  ( ,b) = A( ,H)(  -m) 2H -  b 2-2H = 0 where, A( ,H) = [2  2 H 2H (1-H) 2-2H ] -1 Gives (?):  * = m[1-  2 H] -1 b * =

Design problem We have the figure power  b LBAP curve

Aggregated TCP flows (work in progess) TCP remains the dominant traffic protocol through all hours of the day. TCP traffic typically contributes over 95% of the total traffic volume (UDP protocol at about 2-5% ). A mixture of both well-known applications (http, ftp, smtp, nntp) (70-80% of the total traffic volume), and less known, contribute significan portions to the TCP traffic mix. Available trace files (NLANR: National Laboratory for Networking Traffic)

Aggregated TCP flows, Cont. Friendly TCP Traffic for Differentiated Services; TCP shaping is effective tool for managing TCP traffic; to decrease the burstiness of UDP and TCP traffic, thereby decreasing the load on the router and switch buffers as well as the latency jitter caused by long queues. -The bursts are spread out over time rather than occur at one time. -Each source is at a higher congestion window when the first drop for that particular source occurs as compared to the case of without shaping. Consequently each source gets a higher throughput and higher goodput (also a higher number of drops). The link utilization is better in this case since all sources do not cut down at around the same instant but rather the drops are spread out over time.

Many thanks to Gregorio Procissi