Modles de performance et mulation pour le dimensionnement autonome - - PowerPoint PPT Presentation

▶
mod les de performance et mulation pour le
SMART_READER_LITE
LIVE PREVIEW

Modles de performance et mulation pour le dimensionnement autonome - - PowerPoint PPT Presentation

Nov 20, 2022 211 likes •471 views

Modles de performance et mulation pour le dimensionnement autonome d'applications distribues base de pp composants Introducing Introducing queuing network queuing network- -based based performance awareness performance awareness

slide-1
SLIDE 1

Modèles de performance et émulation pour le dimensionnement autonome d'applications distribuées à base de pp composants Introducing Introducing queuing network queuing network-

  • based

based performance awareness performance awareness in in autonomic systems autonomic systems performance awareness performance awareness in in autonomic systems autonomic systems

(ICAS (ICAS 2010 2010) )

Ahmed Harbaoui, Nabila Salmi Bruno Dillenseger and Jean-Marc Vincent g

Orange Labs Orange Labs, France Telecoms France Telecoms LIG/INRIA LIG/INRIA MESCAL Team MESCAL Team LIG/INRIA LIG/INRIA-MESCAL Team MESCAL Team Grenoble, France

slide-2
SLIDE 2

Outline Outline

1 Towards autonomic management

  • 1. Towards autonomic management
  • 2. Modeling black boxes

3 Automatic black box model identification

  • 3. Automatic black box model identification
  • 4. Experimental results
  • 5. Conclusion & future work

2 Journée ASR Grand-Est Juin 2010 Nancy

slide-3
SLIDE 3

Towards Autonomic management

Decision

QoS ? Performance prediction Autonomic t Candidate configurations SLO (Service Level Objectives) satisfied ? management policies/rules

Modeling + Analysis

Observation Feedback

Self-manage

Distribution Important kl d Complexity workload

QoS loss Autonomic systems

3 Journée ASR Grand-Est Juin 2010 Nancy

slide-4
SLIDE 4

Overview of our global approach

Ready to use Ready-to-use models repository

System Under Test Queuing models

1 2 3 Candidate Several black boxes Service Level Obj ti (SLO) Candidate solutions Choose the solution to apply Performance di ti Objective (SLO) to the system prediction

Queueing networks

4 Journée ASR Grand-Est Juin 2010 Nancy

Queueing networks

slide-5
SLIDE 5

Automatic black box model identification

Goal: Capture the complete behavior of a black box and limit points

Principle

  • 1. Several automatic

load injection steps.

  • 2. Start with a low

injected load, increase j load until black box saturation.

  • 3. Collect measures in

each step (response times resources times, resources utilization), deduce a queueing model

5 Journée ASR Grand-Est Juin 2010 Nancy

slide-6
SLIDE 6

Outline Outline

1 Towards autonomic management

  • 1. Towards autonomic management
  • 2. Modeling black boxes

3 Automatic black box model identification

  • 3. Automatic black box model identification
  • 4. Experimental results
  • 5. Conclusion & future work

6 Journée ASR Grand-Est Juin 2010 Nancy

slide-7
SLIDE 7

Modeling : Modeling :

  • Load

Load-

  • dependent black boxes

dependent black boxes : : Queuing and service times depend on the load.

  • Load

Load-

  • independent black boxes :

independent black boxes : The service time does not depend on the load. depend on the load.

  • Constant delay black boxes :

Constant delay black boxes : Service time does not Service time does not y depend on the load and there is no queuing depend on the load and there is no queuing We define the type of each black box according to the test We define the type of each black box according to the test results results

7 Journée ASR Grand-Est Juin 2010 Nancy

results results

slide-8
SLIDE 8

Queuing models Q g

Queue with 1 server

System Utili ti System Repartition of nb. customers (mem & service) A i l f t ( )

λ

Utilization rate Departure of customers Queue Service μ (Z) Response time (wait+service) Arrival of customers ( )

λ

λ = arrival rate of customers : mean number of arrivals per time unit

ser ice rate : mean b f d t ti it

μ

(wait+service)

= service rate : mean number of served customers per time unit

μ

Z = scheduling policy : FIFO, PS, RR, random, …

Queue model : / X / K T / m / Z Queue model : / X / K T

Interarrival time distribution Service time distribution Q C i

/ m / Z

8 Journée ASR Grand-Est Juin 2010 Nancy

Number of servers Queue Capacity

slide-9
SLIDE 9

Queueing models

M/M/K model : Inter-arrivals Exponential(λ), Service Exponential(μ), Infinite Capacity, K servers, FIFO Analyzable by MVA algorithm Exponential distribution (λ)

Prob {T t}=

≤

≥ t , 1

t

e μ

−

−

M/G/K model : Inter-arrivals Exponential(λ), General service, Infinite capacity, K servers, FIFO Analyzable by R. Marie Algorithm G/G/K model : General Inter-arrivals, General service, Infinite capacity, K servers, FIFO y y g Simulable

9 Journée ASR Grand-Est Juin 2010 Nancy

, Simulable

slide-10
SLIDE 10

Java Modelling Tool (JMT) g ( )

Suite of tools developed by Politecnico di Milano, 2006 -2009 Si J li ti Six Java applications:

  • 1. JSIMGraph, JSIMWiz: QN models

designer/simulator (graphical/ Wizard interface)

  • 2. JMVA: Mean Value Analysis of BCMP compliant

QN models Q

  • 3. JABA: asymptotic analysis of QN models for

identification of the bottlenecks identification of the bottlenecks

  • 4. JWAT: Workload Analysis from log/ used data
  • 5. JMCH : Markov chain (M/M/1, M/M/1/K models)

simulator

10 Journée ASR Grand-Est Juin 2010 Nancy

slide-11
SLIDE 11

Inter-arrival times distribution

Collect arrival times of submitted requests Deduce the inter-arrival sample, Identify the shape of the inter-arrival sample

se statistical tests against se eral distrib tion families e ponential

use statistical tests against several distribution families: exponential,

heavy-tail, etc (Kolmogorov-Smirnov test).

Keep distributions whose p-value > 0.1 Estimate distribution parameters with the Maximum likelihood estimator

method. Realistic requests

11 Journée ASR Grand-Est Juin 2010 Nancy

slide-12
SLIDE 12

Service time distribution

Inject load requests with exponential inter-arrivals. Collect response times (Rk) inter-arrival times (tk) and utilization Collect response times (Rk), inter arrival times (tk) and utilization

  • f all resources (U).

Inferring service times (Xk)1≤k≤n

1 server Several servers

Identify the shape of the service time sample with Kolmogorov-Smirnov

tests tests.

Validation : Compare empirical measures with theoretical ones p p

mean response time, mean waiting time bound

12 Journée ASR Grand-Est Juin 2010 Nancy

slide-13
SLIDE 13

Number of servers (K) Number of servers (K)

  • Cmax= Cmax0 , K=1
  • Iterate until saturation

– When reaching the maximal load Cmax, check the utilization of all black box resources all black box resources. If , for all resource, U < 1 K=K+1; Cmax= Cmax0*K

– If, for a resource, U ≈ 1 If, for a resource, U 1

Stop injection experiment, K = last identified value

13 Journée ASR Grand-Est Juin 2010 Nancy

slide-14
SLIDE 14

Achieving self-regulated injection g g j

  • 1. Injection policy
  • Initial maximal load Cmax = 1/ R

(R: response time measures sample)

  • Injection step, Rising period
  • Sampling period: Number of measures n >=
  • 2. Estimation of the stabilization time
  • 2. Estimation of the stabilization time

Stabilization time = convergence time of the queue Markov chain

(Restriction to Engset models) (Restriction to Engset models)

14 Journée ASR Grand-Est Juin 2010 Nancy

slide-15
SLIDE 15

Outline Outline

1 Towards autonomic management

  • 1. Towards autonomic management
  • 2. Modeling black boxes

3 Automatic black box model identification

  • 3. Automatic black box model identification
  • 4. Experimental results
  • 5. Conclusion & future work

15 Journée ASR Grand-Est Juin 2010 Nancy

slide-16
SLIDE 16

CLIF Load Injection Framework CLIF Load Injection Framework

Probes measure usage of arbitrary computing resources load injector 1 arbitrary computing resources load injector 2 probes arbitrary system under test (SUT) test supervision probes load injector 2 load injector n Load injectors : Execution control and monitoring of Load injectors :

  • send requests, wait for replies, measure response times

g load injectors and resource probes.

  • according to a given scenario
  • for example, emulating the load of a number of real users

16 Journée ASR Grand-Est Juin 2010 Nancy

(through so-called virtual users)

slide-17
SLIDE 17

Architecture of FAMI

(Framework for Automatic Modelling Identification) (Framework for Automatic Modelling Identification)

Model T/X/k :

  • Interarrival distribution
  • Service times distribution
  • Service times distribution
  • Number of Servers

17 Journée ASR Grand-Est Juin 2010 Nancy

slide-18
SLIDE 18

Outline Outline

1 Towards autonomic management

  • 1. Towards autonomic management
  • 2. Modeling black boxes

3 Automatic black box model identification

  • 3. Automatic black box model identification
  • 4. Experimental results
  • 5. Conclusion & future work

18 Journée ASR Grand-Est Juin 2010 Nancy

slide-19
SLIDE 19

Experimental results p

  • Test-bed: Rubis Web-based application, workstation with 2

pp , PIII 1.4Ghz, 1GO RAM.

  • Injector Machine: workstation with quadri-processor Xeon

2Gh 2GO RAM 2Ghz, 2GO RAM.

  • 1st injection step
  • 1st injection step

– Average service time: X0 = 0.021 s

– Theoretical stabilization time = 0.043s

– Cmax0 = 45.30 Virtual users (requests)/s

After 27 min tes of the e periment (12 steps) e reached

  • After 27 minutes of the experiment (12 steps), we reached
  • Number of virtual users = 120 vusers
  • Saturated resource : CPU CPU load = 96%

19 Journée ASR Grand-Est Juin 2010 Nancy

slide-20
SLIDE 20

Experimental results

Inter-arrival times distribution

CV2=1 015

with a 95% confidence interval

CV2=1.015

with a 95% confidence interval

cv2 a possible fitness to an exponential distribution λ =30 34 req/s λ 30.34 req/s Kolmogorov-Smirnov test : p-value=0.59

20 Journée ASR Grand-Est Juin 2010 Nancy

slide-21
SLIDE 21

Experimental results

Service times distribution

100 Vusers 120 Vusers 80 Vusers

Identified distributions: Exponential (M), Lognormal (LN), Hyperexponential (Hr), Gamma (Г), Weibull (Weib)

21 Journée ASR Grand-Est Juin 2010 Nancy

yp p ( ), ( ), ( )

slide-22
SLIDE 22

Experimental results

Identified Queue models

Light load : M/LN/K model Light load : M/LN/K model Heavy load : M/M/K model

22 Journée ASR Grand-Est Juin 2010 Nancy

slide-23
SLIDE 23

Conclusion Conclusion

Automatic performance modelling process of black boxes, based

  • n a self regulated load injection testing and a theoretical approach.

Benefit: performance prediction of an assembly of black boxes, useful when applying autonomic features. Development of a framework, FAMI, based on CLIF, which delivers a set of queue models for a range of workload. q g Difficulties:

Isolating a black box. Configuration options necessary to test a black box ( maximum connections, concurrent threads, ...).

23 Journée ASR Grand-Est Juin 2010 Nancy

slide-24
SLIDE 24

Future work Future work

Improve the computation time of the stabilization time. Integrate, in the FAMI framework, a performance analysis/simulation tool. Instantiate the obtained FAMI framework to achieve self sizing Instantiate the obtained FAMI framework to achieve self-sizing feature.

24 Journée ASR Grand-Est Juin 2010 Nancy

slide-25
SLIDE 25

Thank you for your attention y y

25 Journée ASR Grand-Est Juin 2010 Nancy