630 likes | 753 Views
Fundamental Characteristics of Queues with Fluctuating Load. VARUN GUPTA Joint with:. Motivation. Requests. Clients. Server Farm. Motivation. Requests. Clients. Server Farm. Motivation. Requests. Clients. Server Farm. Motivation. Requests. Clients. Server Farm. Motivation.
E N D
Fundamental Characteristics of Queues with Fluctuating Load VARUN GUPTA Joint with:
Motivation Requests Clients Server Farm
Motivation Requests Clients Server Farm
Motivation Requests Clients Server Farm
Motivation Requests Clients Server Farm
Motivation Requests Clients Server Farm
Motivation Requests Clients Server Farm
Motivation Requests Clients Server Farm
Motivation Requests Real World Fluctuating arrival and service intensities Clients Server Farm
A Simple Model exp(H) High Load L H Low Load exp(L)
H H L L A Simple Model exp() High Load H,H L,L Low Load exp() • Poisson Arrivals • Exponential Job Size Distribution • H/H> L/L • H>Hpossible, only need stability
L L 0 1 2 . . . L L H H 0 1 2 . . . H H The Markov Chain Number of jobs L Phase H Solving the Markov chain provides no behavioral insight
H H L L • N = Number of jobs in the fluctuating load system • Lets try approximating N using (simpler) non-fluctuating systems
H H L L Method 1 Nmix
H H L L Method 1 ½ + Nmix ½ Q: Is Nmix≈ N? A: Only when 0 ,
H H L L Method 2
Method 2 avg(H,L) ≡ Navg avg(H,L) Q: Is Navg ≈ N? A: When ,
Example H=1, H=0.99 L=1, L=0.01 0 E[Navg] = 1 E[Nmix] ≈ 49.5
Observations • Fluctuating system can be worse than non-fluctuating • 0 and asymptotes can be very far apart E[Nmix] > E[Navg] E[Nmix] E[Navg]
Questions • Is fluctuation always bad? • Is E[N] monotonic in ? • Is there a simple closed form approximation for E[N] for intermediate ’s? • How do queue lengths during High Load and Low Load phase compare? How do they compare with Navg? More than 40 years of research has not addressed such fundamental questions!
Not covered in this talk Please read paper. Outline • Is E[Nmix] ≥ E[Navg], always? • Is E[N] monotonic in ? • Simple closed form approximation for E[N] • Application: Capacity Planning • Stochastic orderings for the number of jobs seen by an arrival into an H phase, L phase
Prior Work ? But cubic equations have a close form solution… Transforms Matrix Analytic & Spectral Analysis Fluid/Diffusion Approximations • - Clarke • - Neuts • Yechiali and Naor • - P. Harrison • Adan and Kulkarni • - Massey • Newell • Abate, Choudhary, Whitt Numerical Approaches Involves solution of cubic Limiting Behavior Involves solution of cubic
H=1, H=0.99 E[Nmix] > E[Navg] L=1, L=0.01 Asymptotics for E[N] (H<H) E[Nmix] E[N] E[Navg] High fluctuation a (switching rate) Low fluctuation
E[Navg] E[N] E[Nmix] a Asymptotics for E[N] (H<H) E[Nmix] E[N] E[Navg] a Q: Is this behavior possible? A: Yes • Agrees with our example (H= L) • Ross’s conjecture for systems with constant service rate: • “Fluctuation increases mean delay”
E[N] E[N] E[N] a a a Our Results (H-H) > (L-L) (H-H) = (L-L) (H-H) < (L-L) • Define the slacks during L and H as • sL = L - L • sH = H - H
Our Results E[N] E[N] E[N] a a a sH > sL sH = sL sH < sL • Define the slacks during L and H as • sL = L - L • sH = H - H • Not load but slacks determine the response times! KEY IDEA
Not covered in this talk Please read paper. Outline • Is E[Nmix] ≥ E[Navg], always? • Is E[N] monotonic in ? • Simple closed form approximation for E[N] • Application: Capacity Planning • Stochastic orderings for the number of jobs seen by an arrival into an H phase, L phase
Not covered in this talk Please read paper. Outline • Is E[Nmix] ≥ E[Navg], always? No • Is E[N] monotonic in ? • Simple closed form approximation for E[N] • Application: Capacity Planning • Stochastic orderings for the number of jobs seen by an arrival into an H phase, L phase
NL NH exp() H,H L,L exp() Notation • NH: Number of jobs in system during H phase • NL: Number of jobs in system during L phase • N = (NH+NL)/2
f g NL=f(g(NL)) Analysis of E[N] H,H First steps: • Note that it suffices to look at switching points • Express • NL= f(NH) • NH = g(NL) • The problem reduces to finding Pr{NH=0} and Pr{NL=0} L,L NL NH
f g H(L -L)0H+ L(H-H)0L- (L -L)(H-H) A + 2(A -A) A-A (A-A) (A-A) Where 0L = 0H = L(-1)(H-H) H(-1)(L-L) The simple way forward… H,H • Find the root of a cubic (the characteristic matrix polynomial in the Spectral Expansion method) • Express E[N] in terms of E[N] = L,L Difficult to even prove the monotonicity of E[N] wrt using this! NL NH
Our approach (contd.) KEY IDEA • Express the first moment as E[N] = f1()r+f0()(1-r) • r is the root of a (different) cubic • r1 as 0 and r0 as
r 1 0 Monotonicity of E[N] • E[N] = f1()r+f0()(1-r) • r is monotonic in E[N] is monotonic in • The cubic for r has maximum power of as 2
Monotonicity of E[N] • E[N] = f1()r+f0()(1-r) • r is monotonic in E[N] is monotonic in • The cubic for r has maximum power of as 2 r 1 c1 0 • Need at least 3 roots for when r=c1 • but has at most 2 roots
Monotonicity of E[N] • E[N] = f1()r+f0()(1-r) • r is monotonic in E[N] is monotonic in • The cubic for r has maximum power of as 2 r c2 1 0 • Need at least 2 positive roots for when r=c2 • but for r>1 product of roots is negative
Monotonicity of E[N] • E[N] = f1()r+f0()(1-r) • r is monotonic in E[N] is monotonic in • The cubic for r has maximum power of as 2 r 1 0 • E[N] is monotonic in !
Not covered in this talk Please read paper. Outline • Is E[Nmix] ≥ E[Navg], always? No • Is E[N] monotonic in ? • Simple closed form approximation for E[N] • Application: Capacity Planning • Stochastic orderings for the number of jobs seen by an arrival into an H phase, L phase
Not covered in this talk Please read paper. Outline • Is E[Nmix] ≥ E[Navg], always? No • Is E[N] monotonic in ? Yes • Simple closed form approximation for E[N] • Application: Capacity Planning • Stochastic orderings for the number of jobs seen by an arrival into an H phase, L phase
Approximating E[N] KEY IDEA • Express the first moment as E[N] = f1()r+f0()(1-r) • r is the root of a (different) cubic • r1 as 0 and r0 as • Approximate r by the root of a quadratic KEY IDEA
9 Exact Approx. 7 5 3 1 10-5 10-4 10-3 10-2 10-1 100 10 Approximating E[N] E[N] H=L=1, H=0.95, L=0.2
9 Exact Approx. 7 5 3 1 10-5 10-4 10-3 10-2 10-1 100 10 Approximating E[N] E[N] H=L=1, H=0.95, L=0.2
18 Exact Approx. 14 10 6 2 10-2 10-1 100 10 Approximating E[N] E[N] H=L=1, H=1.2, L=0.2
18 Exact Approx. 14 10 6 2 10-2 10-1 100 10 Approximating E[N] E[N] H=L=1, H=1.2, L=0.2
Not covered in this talk Please read paper. Outline • Is E[Nmix] ≥ E[Navg], always? No • Is E[N] monotonic in ? Yes • Simple closed form approximation for E[N] • Application: Capacity Planning • Stochastic orderings for the number of jobs seen by an arrival into an H phase, L phase
Not covered in this talk Please read paper. Outline • Is E[Nmix] ≥ E[Navg], always? No • Is E[N] monotonic in ? Yes • Simple closed form approximation for E[N] • Application: Capacity Planning • Stochastic orderings for the number of jobs seen by an arrival into an H phase, L phase
2H H H H L L 2L L Application: Capacity Provisioning Scenario Aim: To keep the mean response times same
2H H 2H H 2L L 2L L Application: Capacity Provisioning Scenario Question: What is the effect of doubling the arrival and service rates on the mean response time?
What happens to the mean response time when , are doubled in the fluctuating load queue? A: Halves B: Reduces by more than half C: Reduces by less than half D: Remains almost the same
What happens to the mean response time when , are doubled in the fluctuating load queue? A: Halves B: Reduces by more than half C: Reduces by less than half D: Remains almost the same