Performance Evaluation • When: Wed. 2:20pm~5:10pm • Where: Room 107 • Instructor: 周承復 • Office hours: by appointment • www.csie.ntu.edu.tw/~ccf • E-mail: firstname.lastname@example.org • Class web page and TA will be announced by this weekend • TBD
Reading Reference • Reference books: • Queueing systems (vol. I & II); Kleinrock • Analytical Performance Modeling for Computer Systems; Y.C. Tay • Prob., Stochastic process., and Queueing Theory; Randy Nelson. • Combinatorial Optimization: Algorithms and Complexity; Christos. H. Papadimitriou and Kenneth Steiglitz • Paper reading & presentation
Grading • Tentative • Hw & Paper presentation : 35% • class attendance (participation): 5-10% • Midterm (1 or 2 times): 20-25% • Term projects: 35-40% (1-2 members)
Projects • 1st programming project: • Discrete Time Event Simulator (in c, c++) • Term project
Objective • An intro. to techniques and tools • Construct and analyze a model (or by simulations) for computer and communication systems designs. • Example • Internet, wireless communication, operating system, parallel and distributed system, database system, programming languages….
Objective • paper reading • Server: Hardware Architecture, Operating Systems • Network • Database Management • Term Project • Analytic model • Simulation (or Emulation) • Measurement (or Implementation)
Outline • Introduction • Probability review • Discrete Time Event Simulator • Stochastic Processes • Elementary queueing theory • M/M/1, M/M/1 variants • M/G/1, M/G/1 variants, and Priority Queue • Intermediate queueing theory • Bounding techniques • Matrix Geometric solutions: high-dimensional MC
Outline (cont.) • Average Value and Fluid approximation • Open and close system • equilibrium decomposition • bottleneck analysis • flow equivalence • Linear Programming (ILP, or Non-Linear Programming)
Single Server System • A single-server system has a capacity C • R is the average rate of the demanded work • What is the stable condition for this system?
Single Queue vs. Multi-Queue • Consider a system has K servers • For a customer, which design is better? • Each server has its own queue • All K servers share a single queue
System Design • If you want to upgrade your system, you will do • Buying k single servers with each has capacity C • Buying a single sever with capacity kC
Waiting in the Bus Station • The inter-arrival time dist. for a bus is exponential dist. with mean 10 min • John has arrived at the bus station for 5 min, so how long has he waited for till the next bus comes?
Priority Queue • Consider a non-preemptive system and two customer classes A and B, with respective arrival and service rate la, ma, and lb, mb , mb<ma . • If we want to design a system with the avg. delay for a customer is small, which class should have higher priority
TCP Model • J. Padhye, V. Firoiu, D. Towsley, and J. Kurose. Modeling TCP throughput: a simple model and its empirical validation. In Proc. SIGCOMM, 303–314, September 1998. • Renewal theory, prob. Based approach
TCP Model • Vishal Misra, Wei-Bo Gong, and Don Towsley, Stochastic Differential Equation Modeling and Analysis of TCP-Window size Behavior, ACM SIGMETRICS 1999. • Fluid models, Stochastic differential equation
BitTorrent • D. Qiu and R. Srikant. Modeling and performance analysis of BitTorrent-like peer-to-peer networks. In Proc. SIGCOMM, 367–378, 2004. • Fluid Model, steady-state analysis
802.11 • Giuseppe Bianchi, Performance Analysis of the IEEE 802.11 Distributed Coordination Function, Selected Areas in Communications, IEEE Journal on, Vol. 18, No. 3. (March 2000), pp. 535-547, • DTMC based approach
Personal Comm. System • Yi-Bing Lin, Seshadri Mohan, Anthony Noerpel, Queueing Priority Channel Assignment Strategies for PCS Hand-Off and Initial Access,In Vehicular Technology, IEEE Transactions on, Vol. 43, No. 3. (1994), pp. • CTMC, probability, poisson process
P2P streaming • Yipeng Zhou, Dah Ming Chiu, John C.S. Lui, A Simple Model for Analyzing P2P Streaming Protocols, IEEE International Conference on Network Protocols (19 October 2007), pp. 226-235 • probability
Distributed Protocols • I. Gupta. On the design of distributed protocols from differential equations. In Proc.ACM Symposium on Principles of Distributed Computing (PODC), 216–225, July 2004 • Epidemic model : differential eqn.
Database System • P.A.Bernstein,A.Fekete,H.Guo,R.Ramakrishnan,and P. Tamma, Relaxed currency serializability for middle-tier caching and replication. In Proc. ACM SIGMOD Int. Conf. Management of Data, 599–610, June 2006. • Bottlenecks and flow equivalence
Storage System • Q. Zhu, Z. Chen, L.Tan, Y. Zhou, K. Keeton, and J.Wilkes. Hibernator: helping disk arrays sleep through the winter. In Proc. ACMSymp. Operating Systems Principles (SOSP), 39(5):177–190, October 2005 • Integer Linear Programming
Cloud Computing • Resource Allocation and Markets • Online Auctions in IaaS Clouds: Welfare and Profit Maximization with Server CostsXiaoxi Zhang etl. • An Online Auction Framework for Dynamic Resource Provisioning in Cloud ComputingWeijie Shi etl.
BIG Data Analytics • Machine Learning and Crowdsourcing • An Online Learning Approach to Improving the Quality of Crowd-SourcingYang Liu etl. • Learning to Rank: Regret Lower Bound and Efficient AlgorithmsRichard Combes etl.
Social Network • Ads and Information Dissemination • Social Network Monetization via Sponsored Viral MarketingParinya Chalermsook etl. • Collecting, Organizing, and Sharing Pins in Pinterest: Interest-driven or Social-driven? Jinyoung Han etl. • Filter & Follow: Do Social Media Encourage Efficient News Curation?Avner May
Probability Review • Probability • Conditional probability • Statistical independent • Theorem of total Probability • Bayes’ Theorem • Random Variable