9
RSt_CGr___Overlay_Benchmarking___IMDEA___v.2.0_2011.05.18.ppt KOM - Multimedia Communications Lab Prof. Dr.-Ing. Ralf Steinmetz © 2011 author(s) of these slides including research results from the KOM research network and TU Darmstadt. Otherwise it is specified at the respective slide 17. Mai 2011 Perspectives in Benchmarking Overlay Networks for Internet Science IMDEA, Madrid, Spain May, 18 th 2011 Dr.-Ing. Aleksandra Kovecevic System under Test (SUT) Benchmarking-Scenario Metrics + Requirements Evaluation Workload Chord Gnutella XYZ.KOM KOM – Multimedia Communications Lab 2 KOM @ Technische Universität Darmstadt Internet Science ? KOM – Multimedia Communications Lab 3 KOM Research - Goals Source: http://www.sycor-asia.com/opencms/as/products_services/complementary_services/Telecommunication/ Future Internet Seamless Multimedia Communications KOM – Multimedia Communications Lab 4 KOM Research Seamless Multimedia Communications N 1 N 2 R N 4 N 3 N B N B N B N B Context-Aware Communication P2P-Overlay Routing Testbeds Network Coding

IMDEA, Madrid, Spain May, 18 2011 S c i e n c e ? Workload ...events.networks.imdea.org/sites/default/files/R... · KOM – Multimedia Communications Lab 5 H(„my data“) = 3107

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: IMDEA, Madrid, Spain May, 18 2011 S c i e n c e ? Workload ...events.networks.imdea.org/sites/default/files/R... · KOM – Multimedia Communications Lab 5 H(„my data“) = 3107

RSt_CGr___Overlay_Benchmarking___IMDEA___v.2.0_2011.05.18.ppt

KOM - Multimedia Communications LabProf. Dr.-Ing. Ralf Steinmetz

© 2011 author(s) of these slides including research results from the KOM research network and TU Darmstadt. Otherwise it is specified at the respective slide17. Mai 2011

Perspectives in BenchmarkingOverlay Networks for Internet Science

IMDEA, Madrid, SpainMay, 18th 2011

Dr.-Ing. Aleksandra Kovecevic

Systemunder

Test (SUT)

Benchmarking-Scenario

Metrics +Requirements Evaluation

Workload

Chord GnutellaXYZ.KOM

KOM – Multimedia Communications Lab 2

KOM @ Technische Universität Darmstadt

Internet Science ?

KOM – Multimedia Communications Lab 3

KOM Research - Goals

Source: http://www.sycor-asia.com/opencms/as/products_services/complementary_services/Telecommunication/

Future Internet

Seamless Multimedia Communications

KOM – Multimedia Communications Lab 4

KOM ResearchSeamless Multimedia Communications

N1 N2R N4N3

NB NB

NB NB

Context-Aware Communication

P2P-Overlay Routing

Testbeds

Network Coding

Page 2: IMDEA, Madrid, Spain May, 18 2011 S c i e n c e ? Workload ...events.networks.imdea.org/sites/default/files/R... · KOM – Multimedia Communications Lab 5 H(„my data“) = 3107

KOM – Multimedia Communications Lab 5

H(„my data“)

= 3107

2207

7.31.10.25

peer-to-peer.info

12.5.7.31

95.7.6.10

86.8.10.18

planet-lab.orgberkeley.edu

29063485

201116221008709

611

89.11.20.15

?

KOM ResearchTechnologies

� Mobile Communication Technologies

� Multimedia Technologies� Network Security

� Peer-to-Peer Systems

� Sensor Networking Technologies� Service-oriented Computing

�Web Technologies

Technologies

Research in

Network SecurityService-oriented Computing

Web Technologies

Mobile CommunicationTechnologies

Multimedia TechnologiesSensor NetworkingTechnologies

Service-Orientation

Design Paradigm

Service-oriented

Solution Logic

Service-oriented

Architecture

Services

Service

Compositions

Service Inventory

is designed to support

the implementation of

is designed to support the

creation and evolution of

is designed to support

the implementation of

select

from the

is comprised of

standardized

provides specific

principles in

support of

is primarily

distinguished by

can be comprised of

provides a distinct

set of principles that

shape the design of

automate business

processes by

assembling

is comprised of

Peer-to-Peer Systems

Imag

ine-

net-

tech

.com

RSt_CGr___Overlay_Benchmarking___IMDEA___v.2.0_2011.05.18.ppt

KOM - Multimedia Communications LabProf. Dr.-Ing. Ralf Steinmetz

© 2011 author(s) of these slides including research results from the KOM research network and TU Darmstadt. Otherwise it is specified at the respective slide17. Mai 2011

Perspectives in BenchmarkingOverlay Networks for Internet Science

IMDEA, SpainMay, 18th 2011

Dr.-Ing. Aleksandra Kovecevic

Systemunder

Test (SUT)

Benchmarking-Scenario

Metrics +Requirements Evaluation

Workload

Chord GnutellaXYZ.KOM

KOM – Multimedia Communications Lab 7

Internet Science

The workshop objectives are � mostly in discussing

what Internet science is or, perhaps, what it should be.

Is Internet science just a new namefor research � in computer networking and distributed

systems?

The intersection with which discipline � is likely to give Internet science its next

significant breakthrough?

Can Internet science revolutionize online social networking?

� Which specific analytic techniques and experimental methods from the social and natural sciences can enrich the traditional apparatus of Internet technology scientists?

� benchmarking … ?

� What are the right metrics for the expanded problem space?

� What are the concrete ways for Internet technologists to contribute to the other engaged disciplines?

� With the specialization being a common path to success in science, why do we expect the holistic multidisciplinary Internet science to succeed at all?

KOM – Multimedia Communications Lab 8

Outline

Why to benchmark P2P systems?� Motivation

� Overview

How to benchmark P2P systems?� Benchmarking process

� Quality Aspects

� Metrics� Benchmarking Platform

Which P2P system to benchmark?� Example of Benchmarking P2P Overlays for

Networked Virtual Environments

Page 3: IMDEA, Madrid, Spain May, 18 2011 S c i e n c e ? Workload ...events.networks.imdea.org/sites/default/files/R... · KOM – Multimedia Communications Lab 5 H(„my data“) = 3107

KOM – Multimedia Communications Lab 9

Outline

Why to benchmark P2P systems?� Motivation

� Overview

How to benchmark P2P systems?� Benchmarking process

� Quality Aspects

� Metrics� Benchmarking Platform

Which P2P system to benchmark?� Example of Benchmarking P2P Overlays for

Networked Virtual Environments

KOM – Multimedia Communications Lab 10

Designing and Tuning a P2P Overlay Today

Common Difficulties:� Designing often from scratch

� Design phase often does not consideralready existing systems

� Performance and design problems discovered too late

� Trial & error approach2a. (Re)-Design

Requirementsfulfilled

3. Evaluate

1. RequirementAnalysis

if not

Done

Idea

if yes

2b. ParameterTuning

if not

KOM – Multimedia Communications Lab 11

Engineering Approach

Benefits:� Use components

with well examined behavior� Reusing catalogue for future applications

Comparative Evaluationessential for those steps!

2. Analyze existing overlays

Requirementsfulfilled

3. Create catalogue

1. RequirementAnalysis

if not

Done

Idea

5. Design4. Select

appropriate mechanisms

6. Evaluate

if yes

KOM – Multimedia Communications Lab 12

Comparability in Peer-to-Peer Systems: Ideal

[Pastry] “Pastry: Scalable, decentralized object location and routing for large-scale peer-to-peer systems"

[Chord] “Chord: a scalable peer-to-peer lookup protocol for Internet applications"

Source: [Rowstron2001], [Stoica2001]

Page 4: IMDEA, Madrid, Spain May, 18 2011 S c i e n c e ? Workload ...events.networks.imdea.org/sites/default/files/R... · KOM – Multimedia Communications Lab 5 H(„my data“) = 3107

KOM – Multimedia Communications Lab 13

Comparability in Peer-to-Peer Systems: Reality

Comparability not possible:�No common understanding of quality aspects�No commonly used workloads and metrics

KOM – Multimedia Communications Lab 14

DFG Research Unit FOR 733: QuaP2PPhase 1: „Improving Quality of Peer-to-Peer Systems“2006 - 2009

Quality of P2P Systems

Retrievability

Coherence

Consistency

Correctness

PerformanceScalability

Flexibility

Stability

Dependability

Service Provisioning

Overlay Operations

Individual Node

Complete System

IP Infrastructure

Availability

Reliability

Robustness/ Fault tolerance

Integrity

Confidentiality

Authentication

Non-repudiation

TrustValidityEfficiencyAdaptability

Security

Costs

Performance

Costs

Fairness

KOM – Multimedia Communications Lab 15

DFG Research Unit FOR 733: QuaP2PPhase 2: „Benchmarking of Peer-to-Peer Systems“2009 – 2012

Definition:� comparative analysis of different systems offering the same functional interface

against a reference (value) or metric using a set of standardized tests, so-called benchmarks.

Goal:� Compare : to find the system which can deal best with the given workload� Evaluate limits : to explore limits of a system under extreme situations

KOM – Multimedia Communications Lab 16

Systemunder

Test (SUT)

Benchmarking-Scenario

Metrics +Requirements Evaluation

Workload

Benchmarking of P2P Overlays

Chord GnutellaXYZ.KOM

Page 5: IMDEA, Madrid, Spain May, 18 2011 S c i e n c e ? Workload ...events.networks.imdea.org/sites/default/files/R... · KOM – Multimedia Communications Lab 5 H(„my data“) = 3107

KOM – Multimedia Communications Lab 17

Systemunder

Test (SUT)

Benchmarking-Scenario

Metrics +Requirements Evaluation

Workload

Example: Benchmarking Support for Heterogeneity

Load balancing factor

Supporting heterogeneity?

Number of peersNumber of requests per peerPopularity distribution of queried objectsType of request…

Chord Gnutella

“XYZ.KOM is stable up to 50.000 peers and 10 requests per second”

Fulfillment of requirements

Comparison of Systems“Given a certain Workload A Gnutella is more stable then Chord”

XYZ.KOM

KOM – Multimedia Communications Lab 18

Outline

Why to benchmark P2P systems?� Motivation

� Overview

How to benchmark P2P systems?� Benchmarking process

� Quality Aspects

� Metrics� Benchmarking Platform

What P2P system to benchmark?� Example of Benchmarking P2P Overlays for

Networked Virtual Environments

KOM – Multimedia Communications Lab 19

Benchmarking ProcessComponents

1. SUT specifies requirements :� Functional requirements indicating

for which type of systema benchmark can be used

� Non-functional requirements defining interpretation rules which indicate good or how bad the benchmarking results of the benchmarked system are

2. SUT defines the workload to be used for the benchmark

3. SUT specifies a set of metricswhich have to be measured

Workload

Metrics Requirements

P2PQualityAspects

KOM – Multimedia Communications Lab 20

Benchmarking ProcessHierarchy of Evaluating Quality Aspects2006 - 2009

Quality of P2P Systems

Retrievability

Coherence

Consistency

Correctness

PerformanceScalability

Flexibility

Stability

Dependability

Service Provisioning

Overlay Operations

Individual Node

Complete System

IP Infrastructure

Availability

Reliability

Robustness/ Fault tolerance

Integrity

Confidentiality

Authentication

Non-repudiation

TrustValidityEfficiencyAdaptability

Security

Costs

Performance

Costs

Fairness

Page 6: IMDEA, Madrid, Spain May, 18 2011 S c i e n c e ? Workload ...events.networks.imdea.org/sites/default/files/R... · KOM – Multimedia Communications Lab 5 H(„my data“) = 3107

KOM – Multimedia Communications Lab 21

Benchmarking ProcessHierarchy of Evaluating Quality Aspects2009 - 2012

Scalability

Stability

Robustness

Validity

Performance

Costs

Efficiency

Fairness

Success ratio

Completeness

Lookup delay

Number of hops

Battery consumption

Relative bandwidth usage

Join/Leave delay

∆ of validity/performance/costs

Atomic / Basic metricsDerived metrics

Recovery time

∆ of validity/performance/costs

KOM – Multimedia Communications Lab 22

Benchmarking ProcessChoice of Metrics

Example: Metric for costs in search overlays� In structured overlays, e.g. DHT:

� Number of hops to reach the destination� In unstructured overlays:

� Number of peers that received search message

�We need COMPARABLE quantification of quality aspects

Request

ResponseSpecificOverlay

KOM – Multimedia Communications Lab 23

Benchmarking ProcessMetrics for Benchmarking P2P Systems

Micro metrics:� Explains the results of macro metrics� Crucial in the research analysis

Example:� Metric for a load distribution of an

overlay

� Measures internal behavior of a mechanism

Macro metrics:� On the application layer� Relevant for assessment of quality of a

mechanism

Example:� Recall, response time, freshness, used

traffic (in Bytes/sec)

� Conclusion independent of a specific overlay

P2P-Search-Overlay

P2P-Search-Overlay

Request

Response

Request

Response

SpecificOverlay

KOM – Multimedia Communications Lab 24

Benchmarking ProcessExample of Evaluating Robustness and Stability

Experimenttimeline

Regularchurn

Measurements

Extreme churn

= e.g. failure of 50% of superpeersor tree branch failure

Joining Queries

Recovery time

Pref1

Pref2

Time

Perform

ance

Performance

variation

Robustness

Stability

Workload Metric

Frequent queriesMassive leaves

Massive failures

Performance / Validity / Costs (P/V/C)VARIATION

Recovery timeP/V/C Variation

Page 7: IMDEA, Madrid, Spain May, 18 2011 S c i e n c e ? Workload ...events.networks.imdea.org/sites/default/files/R... · KOM – Multimedia Communications Lab 5 H(„my data“) = 3107

KOM – Multimedia Communications Lab 25

Benchmarking Process

System

under

Test (SUT)

Benchmarking-Scenario

Metrics +

RequirementsEvaluation

Workload

ScenariosScenarios

MeasurementsMeasurements ModelingModeling Workload GeneratorWorkload Generator BenchmarkingPlatform

BenchmarkingPlatform

Simulation EngineSimulation Engine

MetricsMetrics

Quality AspectsQuality Aspects

BenchmarkDescriptionBenchmarkDescription

ApplicationsApplications

input forderived from

RequirementsRequirements

Application ModelApplication Model

Churn ModelChurn Model

User ModelUser Model

Service ModelService Model

KOM – Multimedia Communications Lab 27

DFG Research Unit FOR 733: QuaP2P

Quality Aspects of P2P Systems

P2P Benchmarking Methodology

Quality of P2P Search Overlays

P2P Search Overlay Layer

Quality of P2P Document and System Management

P2P Service Layer

Quality of P2P-Management Mechanisms

P2P Application Layer

P2P Benchmarking Platform

Measurements

P2P-Gaming, Social Knowledge Network, First Response Scenario

Workloads

N. Liebau R. Steinmetz K. Wehrle

W. Effelsberg

A. Buchmann

M. Mühlhäuser T. Strufe

A. Schürr R. Steinmetz

KOM – Multimedia Communications Lab 28

Outline

Why to benchmark P2P systems?� Motivation

� Overview

How to benchmark P2P systems?� Benchmarking process

� Quality Aspects

� Metrics� Benchmarking Platform

Which P2P system to benchmark?� Example of Benchmarking P2P Overlays for

Networked Virtual Environments

KOM – Multimedia Communications Lab 29

Benchmarking Overlays for Networked Virtual Environments (NVE)

Systemunder

Test (SUT)

Benchmarking-Scenario

Metrics +Requirements Evaluation

Workload

Page 8: IMDEA, Madrid, Spain May, 18 2011 S c i e n c e ? Workload ...events.networks.imdea.org/sites/default/files/R... · KOM – Multimedia Communications Lab 5 H(„my data“) = 3107

KOM – Multimedia Communications Lab 30

Characteristics and Requirements for Networked Virtual Environments

Properties /characteristics:� Each participant has a limited vision and

interaction range = area of interest (AOI)� All virtual objects and participants have

spatial and temporal alignment in the virtual space

� Our focus : spatial information dissemination where participants move in a continuous virtual world Requirements:

� Consistency of spatial placement of virtual objects and participants

� Real-time behavior, requiring a high responsiveness and low update delays

� Efficient queries for objects and/or disseminating data in a specific region of the virtual world

Objects

Participants

Legend:

AOI (Area of Interest )

KOM – Multimedia Communications Lab 32

Benchmarking Methodology: Workload

A workload should:� reflect the most critical situations within

the lifetime of a SUT � drive SUT to its performance limits to

gain insights on performance bottlenecks

Our workloads:

time

#pee

rs

Heavy churn

time

#pee

rs

Linearly increasing density

time

node

spe

ed

Linear increasing node speed

time

#pee

rs

Massive join

time

#pee

rs

Massive leave/crash

time

#los

t msg

s

Message loss

*Assumption: fixed world size and AOI for each peer

Density ↑�

# AOI neighbors ↑�

bandwidth utilization ↑

* Fixed world and AOI size

Speed ↑�neighborhood change rapidly AND

effort to maintain neighborhood ↑

exponential churn model with a decreasing mean session length

requires reorganization of the neighborhood relations of a large set of peers within a short amount of time

loss of position updates �inconsistent view of peers OR collapse of overlay topology

Systemunder

Test (SUT)

Benchmarking-Scenario

Metrics +Requirements Evaluation

Workload

Baseline

Baseline

Baseline workload:typical average workload

KOM – Multimedia Communications Lab 33

Contact us…

http://www.quap2p.de

http://research.spec.org

http://peerfactsim.com

KOM – Multimedia Communications Lab 34

Some References

Stoica, I., R Morris, D Karger, F Kaashoek, and H. Balakrishnan. Chord: Scalable Peer-to-Peer Lookup Service for Internet Applications. In Proceedings SIGCOMM, ACM, 2001.

[Stoica2001]

Rowstron, A., and P. Druschel. Pastry: Scalable, decentralized object location, and routing for large-scale peer-to-peer systems. In International Conference on Distributed Systems Platforms, Springer, 2001.

[Rowstron2001]

Gross C., Lehn M., Münker C.,Towards a Common Performance Evaluation of Overlaysfor Networked Virtual Environments, In Eleventh International Conference on Peer-to-Peer Computing, IEEE, 2011, (under submission).

[Gross2011]

Kovacevic, A., Peer-to-Peer Location-based Search: Engineering a Novel Peer-to-Peer Overlay Network, PhD thesis, Technische Universität Darmstadt, 2009.

[Kovacevic2009]

Bharambe, A.R., S. Rao, and S. Seshan, Mercury: a scalable publish-subscribe system for internet games. In Proceedings of the 1st workshop on Network and system support for games, ACM, 2002

[Bharambe2002]

Schmieg, Arne, Michael Stieler, Sebastian Jeckel, Patric Kabus, Bettina Kemme, and Alejandro Buchmann. pSense-Maintaining a Dynamic Localized Peer-to-Peer Structure for Position Based Multicast in Games. In Eighth International Conference on Peer-to-Peer Computing, IEEE, 2008.

[Schmieg 2008]

Hu, S., Tsu-han C. VON: A Scalable Peer-to-Peer Network for Virtual Environments. In IEEE Network, 2006

[Hu2006]

Page 9: IMDEA, Madrid, Spain May, 18 2011 S c i e n c e ? Workload ...events.networks.imdea.org/sites/default/files/R... · KOM – Multimedia Communications Lab 5 H(„my data“) = 3107

KOM – Multimedia Communications Lab 35

Questions & Contact