Systems with unlimited supply of work: MCQN with infinite virtual buffers A Push Pull multiclass system

Σχετικά έγγραφα
ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 19/5/2007

Statistical Inference I Locally most powerful tests

2 Composition. Invertible Mappings

Other Test Constructions: Likelihood Ratio & Bayes Tests

The challenges of non-stable predicates

Phys460.nb Solution for the t-dependent Schrodinger s equation How did we find the solution? (not required)

Homework 8 Model Solution Section

Section 8.3 Trigonometric Equations

Homework 3 Solutions

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 6/5/2006

EE512: Error Control Coding

Lecture 2: Dirac notation and a review of linear algebra Read Sakurai chapter 1, Baym chatper 3

5.4 The Poisson Distribution.

Partial Differential Equations in Biology The boundary element method. March 26, 2013

HOMEWORK 4 = G. In order to plot the stress versus the stretch we define a normalized stretch:

Fractional Colorings and Zykov Products of graphs

Approximation of distance between locations on earth given by latitude and longitude

Fourier Series. MATH 211, Calculus II. J. Robert Buchanan. Spring Department of Mathematics

C.S. 430 Assignment 6, Sample Solutions

Μηχανική Μάθηση Hypothesis Testing

Areas and Lengths in Polar Coordinates

Math 6 SL Probability Distributions Practice Test Mark Scheme

DESIGN OF MACHINERY SOLUTION MANUAL h in h 4 0.

ST5224: Advanced Statistical Theory II

Matrices and Determinants

Bayesian statistics. DS GA 1002 Probability and Statistics for Data Science.

Jesse Maassen and Mark Lundstrom Purdue University November 25, 2013

ΕΛΛΗΝΙΚΗ ΔΗΜΟΚΡΑΤΙΑ ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΡΗΤΗΣ. Ψηφιακή Οικονομία. Διάλεξη 10η: Basics of Game Theory part 2 Mαρίνα Μπιτσάκη Τμήμα Επιστήμης Υπολογιστών

Depth versus Rigidity in the Design of International Trade Agreements. Leslie Johns

ΗΜΥ 220: ΣΗΜΑΤΑ ΚΑΙ ΣΥΣΤΗΜΑΤΑ Ι Ακαδημαϊκό έτος Εαρινό Εξάμηνο Κατ οίκον εργασία αρ. 2

Nowhere-zero flows Let be a digraph, Abelian group. A Γ-circulation in is a mapping : such that, where, and : tail in X, head in

the total number of electrons passing through the lamp.

Estimation for ARMA Processes with Stable Noise. Matt Calder & Richard A. Davis Colorado State University

Figure A.2: MPC and MPCP Age Profiles (estimating ρ, ρ = 2, φ = 0.03)..

Exercises 10. Find a fundamental matrix of the given system of equations. Also find the fundamental matrix Φ(t) satisfying Φ(0) = I. 1.

SCHOOL OF MATHEMATICAL SCIENCES G11LMA Linear Mathematics Examination Solutions

Areas and Lengths in Polar Coordinates

Problem Set 3: Solutions

The Simply Typed Lambda Calculus

Second Order Partial Differential Equations

Main source: "Discrete-time systems and computer control" by Α. ΣΚΟΔΡΑΣ ΨΗΦΙΑΚΟΣ ΕΛΕΓΧΟΣ ΔΙΑΛΕΞΗ 4 ΔΙΑΦΑΝΕΙΑ 1

Exercises to Statistics of Material Fatigue No. 5

Assalamu `alaikum wr. wb.

3.4 SUM AND DIFFERENCE FORMULAS. NOTE: cos(α+β) cos α + cos β cos(α-β) cos α -cos β

Απόκριση σε Μοναδιαία Ωστική Δύναμη (Unit Impulse) Απόκριση σε Δυνάμεις Αυθαίρετα Μεταβαλλόμενες με το Χρόνο. Απόστολος Σ.

TMA4115 Matematikk 3

Econ 2110: Fall 2008 Suggested Solutions to Problem Set 8 questions or comments to Dan Fetter 1

Models for Probabilistic Programs with an Adversary

ΕΛΛΗΝΙΚΗ ΔΗΜΟΚΡΑΤΙΑ ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΡΗΤΗΣ. Ψηφιακή Οικονομία. Διάλεξη 8η: Producer Behavior Mαρίνα Μπιτσάκη Τμήμα Επιστήμης Υπολογιστών

Instruction Execution Times

Πρόβλημα 1: Αναζήτηση Ελάχιστης/Μέγιστης Τιμής

Second Order RLC Filters

[1] P Q. Fig. 3.1

Numerical Analysis FMN011

UNIVERSITY OF CALIFORNIA. EECS 150 Fall ) You are implementing an 4:1 Multiplexer that has the following specifications:

Overview. Transition Semantics. Configurations and the transition relation. Executions and computation

4.6 Autoregressive Moving Average Model ARMA(1,1)

Forced Pendulum Numerical approach

Capacitors - Capacitance, Charge and Potential Difference

Tridiagonal matrices. Gérard MEURANT. October, 2008

( y) Partial Differential Equations

CHAPTER 25 SOLVING EQUATIONS BY ITERATIVE METHODS

Math221: HW# 1 solutions

"ΦΟΡΟΛΟΓΙΑ ΕΙΣΟΔΗΜΑΤΟΣ ΕΤΑΙΡΕΙΩΝ ΣΥΓΚΡΙΤΙΚΑ ΓΙΑ ΤΑ ΟΙΚΟΝΟΜΙΚΑ ΕΤΗ "

Higher Derivative Gravity Theories

Example Sheet 3 Solutions

ΕΛΛΗΝΙΚΗ ΔΗΜΟΚΡΑΤΙΑ ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΡΗΤΗΣ. Ψηφιακή Οικονομία. Διάλεξη 7η: Consumer Behavior Mαρίνα Μπιτσάκη Τμήμα Επιστήμης Υπολογιστών

AKAΔΗΜΙΑ ΕΜΠΟΡΙΚΟΥ ΝΑΥΤΙΚΟΥ ΜΑΚΕΔΟΝΙΑΣ ΣΧΟΛΗ ΜΗΧΑΝΙΚΩΝ ΠΤΥΧΙΑΚΗ ΕΡΓΑΣΙΑ ΘΕΜΑ: Η ΧΡΗΣΗ ΒΙΟΚΑΥΣΙΜΩΝ ΣΤΗΝ ΝΑΥΤΙΛΙΑ ΠΛΕΟΝΕΚΤΗΜΑΤΑ-ΜΕΙΟΝΕΚΤΗΜΑΤΑ ΠΡΟΟΠΤΙΚΕΣ

Every set of first-order formulas is equivalent to an independent set

Dynamic types, Lambda calculus machines Section and Practice Problems Apr 21 22, 2016

(1) Describe the process by which mercury atoms become excited in a fluorescent tube (3)

CYTA Cloud Server Set Up Instructions

Lecture 2. Soundness and completeness of propositional logic

Chapter 6: Systems of Linear Differential. be continuous functions on the interval

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 24/3/2007

ECE 308 SIGNALS AND SYSTEMS FALL 2017 Answers to selected problems on prior years examinations

Πανεπιστήµιο Κύπρου Πολυτεχνική Σχολή

Concrete Mathematics Exercises from 30 September 2016

Inverse trigonometric functions & General Solution of Trigonometric Equations

Queueing Theory I. Summary. Little s Law Queueing System Notation Stationary Analysis of Elementary Queueing Systems. M/M/1 M/M/m M/M/1/K

Practice Exam 2. Conceptual Questions. 1. State a Basic identity and then verify it. (a) Identity: Solution: One identity is csc(θ) = 1

On a four-dimensional hyperbolic manifold with finite volume

ΚΥΠΡΙΑΚΟΣ ΣΥΝΔΕΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY 21 ος ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ Δεύτερος Γύρος - 30 Μαρτίου 2011

Reminders: linear functions

General 2 2 PT -Symmetric Matrices and Jordan Blocks 1

Potential Dividers. 46 minutes. 46 marks. Page 1 of 11

Managing Economic Fluctuations. Managing Macroeconomic Fluctuations 1

ω ω ω ω ω ω+2 ω ω+2 + ω ω ω ω+2 + ω ω+1 ω ω+2 2 ω ω ω ω ω ω ω ω+1 ω ω2 ω ω2 + ω ω ω2 + ω ω ω ω2 + ω ω+1 ω ω2 + ω ω+1 + ω ω ω ω2 + ω

derivation of the Laplacian from rectangular to spherical coordinates

Srednicki Chapter 55

Ψηφιακή Οικονομία. Διάλεξη 11η: Markets and Strategic Interaction in Networks Mαρίνα Μπιτσάκη Τμήμα Επιστήμης Υπολογιστών

New bounds for spherical two-distance sets and equiangular lines

Monetary Policy Design in the Basic New Keynesian Model


PARTIAL NOTES for 6.1 Trigonometric Identities

Solutions to Exercise Sheet 5

Notes on the Open Economy

ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΥΠΡΟΥ ΤΜΗΜΑ ΠΛΗΡΟΦΟΡΙΚΗΣ. ΕΠΛ342: Βάσεις Δεδομένων. Χειμερινό Εξάμηνο Φροντιστήριο 10 ΛΥΣΕΙΣ. Επερωτήσεις SQL

Physical DB Design. B-Trees Index files can become quite large for large main files Indices on index files are possible.

Transcript:

Systems with unlimited supply of work: MCQN with infinite virtual buffers A Push Pull multiclass system Gideon Weiss University of Haifa Joint work with students: Anat (Anastasia) Kopzon Yoni Nazarathy

Queue vs Manufacturing machine: Single server queue max(,) 1

Queue vs Manufacturing machine: Single server queue max(,) Machine with controlled input

Infinite supply of work - infinite virtual buffers A tandem of queues 1 3

Infinite supply of work - infinite virtual buffers A tandem of queues 1 3 4

Infinite supply of work - infinite virtual buffers A tandem of queues 1 3 The push pull system 1 1 5

Balanced full utilization 1 1 1 ν 1 ν Both machines work all time and no flow accumulates implies: ν 1 = α 1 1 = (1 α ) 1 ν = α = (1 α 1 ) 6

Balanced full utilization 1 1 1 ν 1 ν ν 1 = 1 1 ( ) 1 1 ν = ( 1 1 ) 1 1 7

Balanced full utilization 1 1 1 ν 1 ν ν 1 = 1 1 ( ) 1 1 ν = ( 1 1 ) 1 1 How does it behave? 8

The Rybko Stolyar network 1 α 1 1 1 α Traffic intensity/offered load ρ 1 = α 1 1 + α ρ = α + α 1 1 9

The Rybko Stolyar network 1 α 1 1 1 α Traffic intensity ρ 1 = α 1 1 + α ρ = α + α 1 1 Heavy traffic: α 1, α ρ 1, ρ Balanced heavy traffic: α 1 ν 1, α ν 10

Push pull system - inherently stable case 1 1 1 ν 1 1 < 1 ν < 11

Push pull system - inherently stable case 1 1 1 ν 1 1 < 1 ν < 0,4 Last buffer first serve, priority to pull over push 0,3 0, 0,1 1 1 1 1 1 1 0,0 1,0,0 3,0 4,0 5,0 1 1 1 1 1 1 1

Push pull system - inherently stable case 1 1 1 ν 1 1 < 1 ν < 0,4 0,3 0, 0,1 1 P(, ) = 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0,0 1,0,0 3,0 4,0 5,0 1 1 1 1 1 1 1 1 1 1 1 1 n 1, = 0 n, = 0 13

Push pull system - inherently stable case 1 1 1 ν 1 1 < 1 ν < 0,4 Sample path consists of randomly switching between M/M/1 periods of the two streams 0,3 0, 0,1 P(, ) = 1 1 1 1 1 1 1 1 1 n 1 n, = 0 1 1 1 1 1 1 0,0 1,0,0 3,0 4,0 5,0 1 1 1 1 1 1 14

Rybko Stolyar network - LBFS virtual machine α 1 1 1 1 α Under LBFS the two pulling queues form a virtual machine - only one works at any time Conditions for stability (global stability of all work conserving policies) ρ 1 = α 1 1 + α ρ = α + α 1 1 virtual machine load= α 1 1 + α < 1 When i > i, we can have ρ 1, ρ < 1 but virtual machine load > 1 LBFS will be unstable 15

Push pull system - inherently unstable case 1 1 1 ν 1 1 > 1 ν > 0,4 Last buffer first serve, priority to pull over push: All the states are transient 0,3 0, 0,1 1 1 1 1 1 1 0,0 1,0,0 3,0 4,0 5,0 1 1 1 1 1 1 16

Push pull system - fixed threshold policy 1 1 1 ν 1 1 > 1 ν > 0,5 0,4 0,3 0, s 1 1 1 1 1,5,5 4,5 1 1 1 1 1,4,4 1 1 1 1 1 1 1 1 1,3,3 3,3 4,3 5,3 1 1, 3, 4, 5, Machine i : Monitor queue at Machine j, if < s j Push, if > s j Pull. s s 1 : s 1 1 < 1 1 0,1,1 3,1 4,1 5,1 s : s 1 1 < 1 0,0 1,0,0 3,0 4,0 5,0 1 1 1 1 1 1 17

18 Push pull system - fixed threshold Steady State: Push pull system - fixed threshold Steady State: Symmetric streams m,0 1 1 1 1 m,1 m,s... P m,n = P s,s n + n 1 s + s 1 m = s,0 n s P s,s + s 1 m s 1 s + s 1 m s+1 n+1 + s m > s,0 n s

Maximum pressure policy Maximum pressure policy will stabilize any system with offered load <1 Consider MCQN with fluid dynamics described by d Q(t) = α Ru(t) dt α where R is the input output matrix, u(t) is the machine allocation, and is the input rate. The machine allocations (controls ) are subject to resource constraints Max pressure attempts to maximize the gradient of the sum of squares of queue lengths d dt d Q k (t) = Q dt (t)q(t) = Q (t)(α Ru(t)) k At any time t choose allocation u(t) such that max Q (t)ru(t) s.t. Au(t) 1, u(t) 0, u(t) is available In balanced heavy traffic it optimizes the diffusion approximation 19

Decisions for max Q (t)ru(t) Rybko-Stolyar α 1 Q 11 (t) Q 1 (t) 1 1 1 Machine 1 Push if Q 11 (t) Q 1 (t) < Q (t) Pull if Q 11 (t) Q 1 (t) > Q (t) Q (t) Q 1 (t) α Machine Push if Q 1 (t) Q (t) < Q 1 (t) Pull if Q 1 (t) Q (t) > Q 1 (t) Infinite supply push pull ν t D 11 (t) Q 1 (t) 1 1 1 ν Q (t) ν 1 ν t D 11 (t) Machine 1 Machine Push if ν 1 t D 11 (t) Q 1 (t) < Q (t) Pull if ν 1 t D 11 (t) Q 1 (t) > Q (t) Push if ν t D 1 (t) Q (t) < Q 1 (t) Pull if ν t D 1 (t) Q (t) > Q 1 (t) 0

Rybko Stolyar network under max pressure = 1.5 ρ = 0.9 ρ = 0.99 ρ = 1.0 ρ = 1. 1

Push Pull network under max pressure ρ = 0.9 ρ = 0.99 = 1.5 ρ = 1.0 ρ = 1.

Push Pull under max pressure - trying for full utilization Sample of 4 more runs = 1.5 ρ = 1.0 3

How good is max pressure? Rybko-Stolyar Infinite supply push pull Q 11 (t) Q 1 (t) 1 ν t D 11 (t) Q 1 (t) 1 α 1 1 1 1 1 ν 1 α ν Q (t) Q 1 (t) Q (t) ν t D 11 (t) Max pressure attempts to balance the queues, minimize sum of squares For infinite supply it will balance fluctuations in supply with queues Using the fixed threshold, fluctuations in supply may be twice as big, But: Average number in the two queues (with full utilization) E(queues)= 11.86 (s 1 =s =4) 4

Push pull system - balancing diagonal policy ν 0,5 1,5,5 3,5 4,5 5,5 1 ν Symmetric > 0,4 0,3 1,4,4 1,3,3 3,4 4,4 5,4 3,3 4,3 5,3 Machine i : Compare queue to machine j, if other queue longer - Pull, if your queue longer - Push, equal queues pull 0, 1,, 3, 4, 5, 0,1 1,1,1 3,1 4,1 5,1 0,0 1,0,0 3,0 4,0 5,0 Gideon Weiss, University of Haifa, Push Pull, 006 E(queues)= 1.37 = 1.5 (compared to 11.86 For fixed threshold) 5

Push pull balancing diagonal policy - steady state Symmetric > 0,5 1,5,5 3,5 4,5 5,5 0,4 0,3 1,4,4 1,3,3 3,4 4,4 5,4 3,3 m > n P m,n = P 0,0 4,3 5,3 m + m 1 1 m 1 m 1 + m 1 1 P m,0 = P 0,0 n + m 1 + m 1 1 m 1 + m 1 1 m 1 0, 1,, 3, 4, 5, 0,1 0,0 1,1 1,0,1 3,1 4,1 5,1,0 3,0 4,0 5,0 P m,m = P 0,0 + m 1 1 m 1 + m 1 1 m 1 6

Generalized threshold policies We define two monotone threshold curves, above the levels s 1, s and use those for our policy We define and appropriate Lyapunov function, and use Foster Lyapunov criterion to show that all of these are stable. 7

A dynamic programming problem For full utilization, or for a given throughput, find actions to minimize expected queue lengths: Restless bandit indexes? 1 1 1 ν 1 ν Machine 1 choices Machine choices 1 1 1 1 1 1 8

What next? We have looked at a small system that can work at full utilization and not be congested ν 1 1 1 ν 9

What next? We have looked at a small system that can work at full utilization and not be congested Provided we feed it infinite supply of work ν 1 1 1 ν 30

What next? The challenge is to have a large network working at full utilization with no congestion, on a dynamic basis - by online control which assures supply of work at selected points. Leads to fluid optimal fluid control, solved as a continuous linear program ν 1 ν 31

What next? The challenge is to have a large network working at full utilization with no congestion, on a dynamic basis - by online control which assures supply of work at selected points. Leads to fluid optimal fluid control, solved as a continuous linear program 3

Summary We presented a small MCQN with infinite virtual buffers and compared it to the similar Rybko-Stolyar network. The infinite supply of work, infinite virtual buffers, present a new paradigm for MCQN in balanced heavy traffic Maximum pressure policies can be adapted to MCQN w infinite virtual buffers and they achieve pathwise stability under full utilization, similar to the Rybko -Stolyar network. However, at full utilization the system will become congested with null recurrent queues that scale as n to a diffusion The greater controllability of MCQN with virtual infinite buffers allows full utilization of the system in which all the random fluctuations are pushed to the input and output of the system, and all the internal queues are not congested 33