Variability, Randomness and Little s Law

Similar documents
Tian Zheng Department of Statistics Columbia University

an application to HRQoL

Multistage Median Ranked Set Sampling for Estimating the Population Median

UNIT10 PLANE OF REGRESSION

V. Principles of Irreversible Thermodynamics. s = S - S 0 (7.3) s = = - g i, k. "Flux": = da i. "Force": = -Â g a ik k = X i. Â J i X i (7.

1.050 Engineering Mechanics I. Summary of variables/concepts. Lecture 27-37

ECE559VV Project Report

9/12/2013. Microelectronics Circuit Analysis and Design. Modes of Operation. Cross Section of Integrated Circuit npn Transistor

The Schrödinger Equation

Physics 11b Lecture #2. Electric Field Electric Flux Gauss s Law

A Queuing Model for an Automated Workstation Receiving Jobs from an Automated Workstation

Chapter 13 - Universal Gravitation

Department of Physics, Korea University Page 1 of 5

Links in edge-colored graphs

Analysis of Discrete Time Queues (Section 4.6)

Chapter 3 Vector Integral Calculus

Chapter 7 Channel Capacity and Coding

FI 2201 Electromagnetism

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming

Remote Sensing. Remote sensing is a quasi-linear estimation problem. Equation of radiative transfer: ) T B e τ T(z) (z)e τ. τ(z)

Merging to ordered sequences. Efficient (Parallel) Sorting. Merging (cont.)

Physics 1501 Lecture 19

Linear Feature Engineering 11

Hydrological statistics. Hydrological statistics and extremes

The Greatest Deviation Correlation Coefficient and its Geometrical Interpretation

Analysis of Queuing Model for Machine Repairing System with Bernoulli Vacation Schedule

4.4 Continuum Thermomechanics

2/24/2014. The point mass. Impulse for a single collision The impulse of a force is a vector. The Center of Mass. System of particles

Dynamic Programming. Preview. Dynamic Programming. Dynamic Programming. Dynamic Programming (Example: Fibonacci Sequence)

Bayesian Assessment of Availabilities and Unavailabilities of Multistate Monotone Systems

XII. Addition of many identical spins

CSE 546 Midterm Exam, Fall 2014(with Solution)

APPENDIX A Some Linear Algebra

A. Thicknesses and Densities

Chapter Fifiteen. Surfaces Revisited

Measurement of Radiation: Exposure. Purpose. Quantitative description of radiation

Space-time Queuing Theoretic Modeling of Opportunistic Multi-hop Coexisting Wireless Networks With and Without Cooperation

LINEAR MOMENTUM. product of the mass m and the velocity v r of an object r r

If there are k binding constraints at x then re-label these constraints so that they are the first k constraints.

Online-routing on the butterfly network: probabilistic analysis

ˆ SSE SSE q SST R SST R q R R q R R q

The Gaussian classifier. Nuno Vasconcelos ECE Department, UCSD

Communication with AWGN Interference

d 4 x x 170 n 20 R 8 A 200 h S 1 y 5000 x 3240 A 243

Chapter 7 Channel Capacity and Coding

Randomness and Computation

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

The Second Anti-Mathima on Game Theory

Support Vector Machines

Application of Queuing Theory to Waiting Time of Out-Patients in Hospitals.

Section 8.3 Polar Form of Complex Numbers

The Forming Theory and the NC Machining for The Rotary Burs with the Spectral Edge Distribution

VEKTORANALYS FLUX INTEGRAL LINE INTEGRAL. and. Kursvecka 2. Kapitel 4 5. Sidor 29 50

A Method of Reliability Target Setting for Electric Power Distribution Systems Using Data Envelopment Analysis

Coarse-Grain MTCMOS Sleep

Physics 111 Lecture 5 (Walker: 3.3-6) Vectors & Vector Math Motion Vectors Sept. 11, 2009

Splay Trees Handout. Last time we discussed amortized analysis of data structures

Integral Vector Operations and Related Theorems Applications in Mechanics and E&M

On the Throughput of Clustered Photolithography Tools:

Some Approximate Analytical Steady-State Solutions for Cylindrical Fin

to point uphill and to be equal to its maximum value, in which case f s, max = μsfn

CS-433: Simulation and Modeling Modeling and Probability Review

Machine Learning. Spectral Clustering. Lecture 23, April 14, Reading: Eric Xing 1

FUZZY FINITE ELEMENT METHOD

The Ordinary Least Squares (OLS) Estimator

P R. Lecture 4. Theory and Applications of Pattern Recognition. Dept. of Electrical and Computer Engineering /

Detection and Estimation Theory

Large scale magnetic field generation by accelerated particles in galactic medium

A NOTE ON ELASTICITY ESTIMATION OF CENSORED DEMAND

Thermodynamics of solids 4. Statistical thermodynamics and the 3 rd law. Kwangheon Park Kyung Hee University Department of Nuclear Engineering

Chapter 23: Electric Potential

Density Functional Theory I

If there are multiple rxns, use concentrations not conversions. These might occur in combination or by themselves.

a. (All your answers should be in the letter!

ISSN X Reliability of linear and circular consecutive-kout-of-n systems with shock model

Support Vector Machines

Moo-rings. Marina Mooring Optimization. Group 8 Route 64 Brian Siefering Amber Mazooji Kevin McKenney Paul Mingardi Vikram Sahney Kaz Maruyama

N = N t ; t 0. N is the number of claims paid by the

A Tutorial on Multiple Integrals (for Natural Sciences / Computer Sciences Tripos Part IA Maths)

VECTOR MECHANICS FOR ENGINEERS: Vector Mechanics for Engineers: Dynamics. In the current chapter, you will study the motion of systems of particles.

Scalars and Vectors Scalar

Physics 207 Lecture 16

Evaluation for sets of classes

Energy in Closed Systems

Homework 1 Solutions CSE 101 Summer 2017

Δt The textbook chooses to say that the average velocity is

Analysis of Queuing Delay in Multimedia Gateway Call Routing

Continuous Time Markov Chains

TCOM 501: Networking Theory & Fundamentals. Lecture 7 February 25, 2003 Prof. Yannis A. Korilis

arxiv: v2 [stat.me] 26 Jun 2012

For now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results.

8 Baire Category Theorem and Uniform Boundedness

Errors for Linear Systems

AP-C WEP. h. Students should be able to recognize and solve problems that call for application both of conservation of energy and Newton s Laws.

CSC321 Tutorial 9: Review of Boltzmann machines and simulated annealing

xp(x µ) = 0 p(x = 0 µ) + 1 p(x = 1 µ) = µ

Cathy Walker March 5, 2010

Copyright 2017 by Taylor Enterprises, Inc., All Rights Reserved. Adjusted Control Limits for P Charts. Dr. Wayne A. Taylor

Lecture 5 Single factor design and analysis

Using T.O.M to Estimate Parameter of distributions that have not Single Exponential Family

Transcription:

Vaalty, Randomness and Lttle s Law Geoge Leopoulos Lttle s Law Assumptons Any system (poducton system) n whch enttes (pats) ave, spend some tme (pocessng tme + watng) and eventually depat Defntons = (long-un expected) aveage thoughput of the system (pats/unt tme) IP = (long-un expected) aveage wok n pocess (nume of pats n the system) (pats) = (long-un expected) aveage cycle tme (tme spent n the system) (tme unts) IP Lttle s Law: IP IP ΤΗ IP 2

Bottleneck ate, aw pocess tme, ctcal IP Poducton lne (wokstatons (S) n sees) S S 2 S 3 S t aveage pocess tme n S s nume of paallel machnes n S s : aveage (maxmum) pocess ate of S t mn{ }: ottleneck ate aveage pocess ate of slowest S maxmum of the l ne T t : aveage aw pocess tme n the lne mnmum T : ctcal IP IP fo whch a lne wth no vaalty acheves maxmum ( ) wth mnmun ( T ) 3 Bottleneck ate, aw pocess tme, ctcal IP Example : (Balanced lne) t = 2 hs s = = ½ pats/h t 2 = 2 hs s 2 = 2 = ½ pats/h t 3 = 2 hs s 3 = 3 = ½ pats/h t = 2 hs s = = ½ pats/h mn { } mn{.5,.5,.5,.5}.5 pats/hs,, T t 2 2 2 2 8 hs T (.5)(8) pats ( no. of machnes) 2

Bottleneck ate, aw pocess tme, ctcal IP Example 2: (Unalanced lne) t = 2 s = = ½ =.5 t 2 = 5 s 2 = 2 2 = 2/5 =. t 3 = s 3 = 6 3 = 6/ =.6 t = 3 s = 2 = 2/3 =.67 mn{ } mn{.5,.,.6,.67}. pats/h,, T t 2 5 3 2 hs T (.)(2) 8 pats ( no. of machnes) 5 Pefomance as a functon of IP Example : (Balanced lne) Best case pefomance (no vaalty: constant pocess tmes) IP (w) %T = IP/ % 8 /8 =.25 25 2 8 2/8 =.25 5 3 8 3/8 =.375 75 8 /8 =.5 5 25 5/ =.5 6 2 5 6/2 =.5 7 75 7/ =.5 8 6 2 8/6 =.5 6 3

Pefomance as a functon of IP Example : (Balanced lne) Best case pefomance 3 T 2 2 6 8 est T, f w w,f w 2 6 8 2 IP (w) 7 Pefomance as a functon of IP Example : (Balanced lne) Best case pefomance,6,5,,3,2, T est w,f w T, f w 2 6 8 2 IP (w) 8

Pefomance as a functon of IP Example : (Balanced lne) ose case pefomance (hghest vaalty: hghly vaale pocess tmes) Assumpton: Fo IP = w, thee ae 2 types of pats: slow pat (wth pocess tme t s = wt at S ) followed y w fast pats (wth zeo (neglgle) pocess tmes t f = ) w wt Aveage pocess tme n S t w Example: w = : t s s = tt = 2 = 8h hs = 8 + 8 + 8 + 8 = 32 hs! (= T ); = /32 = /8 9 Pefomance as a functon of IP Example : (Balanced lne) ose case pefomance IP (w) t s %T = IP/ % 2 2 = 8 /8 =.25 25 2 = 6 2 2/6 =.25 25 3 6 6 = 2 3 3/2 =.25 25 8 8 = 32 /32 =.25 25 5 = 5 5/ =.25 25 6 2 2 = 8 6 6/8 =.25 25 7 = 56 7 7/56 =.25 25 8 6 6 = 6 8 8/6 =.25 25 5

Pefomance as a functon of IP Example : (Balanced lne) ost case pefomance 3 T 2 2 6 8 T wt wost 2 6 8 2 IP (w) Pefomance as a functon of IP Example : (Balanced lne) Best case pefomance T,6,5,,3,2, wost T 2 6 8 2 IP (w) 2 6

Pefomance as a functon of IP Example : (Balanced lne) Pactcal wose case pefomance (hghest andomness medum vaalty) Fo IP = w, all possle states ae equally lkely Example: w = 3 (2 states) N N2 N3 N 3 3 3 3 2 2 3 Pefomance as a functon of IP Example : (Balanced lne) Pactcal wose case pefomance Assumptons. Lne s alanced 2. All statons have a sngle machne 3. Pocess tmes ae exponentally dstuted Defntons N = nume of sngle-machne wokstatons t = aveage pocessng tme at each S Aveage tme spent at a staton watng tme pocess tme = w tt N w w Nt t Nt ( w) t T N IP w w w T w 7

Pefomance as a functon of IP Example : (Unalanced lne example) Pactcal wost case pefomance T 6 5 3 2 wost PC / Unal. lne 2 est w T PC 8 2 6 2 2 IP 5 Pefomance as a functon of IP Example : (Balanced lne) Pactcal wost case pefomance,5,,3,2, 8 2 6 2 2 IP est Unal. lne 2 wost PC PC w w 6 8