Validation of Service Oriented Computing DEVS Simulation Models Hessam Sarjoughian and Mohammed Muqsith Arizona Center for Integrative Modeling & Simulation School of Computing, Informatics, and Decision Systems Engineering Dazhi Huang and Stephen Yau Information Assurance Center School of Computing, Informatics, and Decision Systems Engineering 1
22
Embed
Validation of Service Oriented Computing DEVS Simulation Models
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Validation of Service Oriented Computing DEVS Simulation Models
Hessam Sarjoughian and Mohammed Muqsith Arizona Center for Integrative Modeling & Simulation School of Computing, Informatics, and Decision Systems Engineering
Dazhi Huang and Stephen Yau Information Assurance Center School of Computing, Informatics, and Decision Systems Engineering
1
2
Motivation A key promise for Service Based Software Systems is
On-demand Quality of Service (QoS)
However, system design with QoS support is challenging QoS depends on
System architecture Interactions among constituent parts under a dynamic environment
Link
Link Link
Voice Communication
Service
EncryptionService
Software Software
Application Application
Software Software
Hardware Hardware
HardwareHardware
Server Server
Client Client
Hub
CommunicationCommunication
3
Motivation Design evaluations addressing QoS in Service Based Software
Systems Difficult to track & inflexible for experimentations Small-scale system QoS can be predicted using analytical
methods Complex interactions in large-scale systems complicate QoS
prediction
Simulation, in contrast to real design and implementation Offers alternate ways of understanding, development, and
experimentation Easier to configure system with repeatable experimentation
capability Early evaluation of system architecture
Simplify complex system design and evaluation Validation is generally a necessity
Link Models physical media Interconnects multiple network
switches Network Switch, Router
Interconnects networks Routes packets
System Service Mapping Provides flexible assignment
of services to processors
Hardware
SOA Compliant
Service
Flexible Map
7
SOC-DEVS: Service Interactions swService accounts for two basic aspects
Operations Denotes functionality provided by the service
Communications Denotes service to service interaction capability
SSM
Operations
Communications
Jobs CPU
Transport UnitMessages
Processor 1NIC
swService 1
Models service to service interaction through hardware layer Jobs
Job (cycles/sec, Mbytes of memory) represents computational load for operations Messages
Message (MsgType, Size) represents communication load for communications
SSM
Jobs
MessagesOperations
Communications
swService 1 Processor M
Jobs
Messages
Router/Switch
swService k
Operations
Communications
CPU
Transport Unit
NIC Link
LinkProcessor NNIC
CPU
Transport Unit
swService accounts for two basic aspects Operations – Denotes functionality provided by the service Communications – Denotes service to service interaction capability
Models service to service interaction through hardware layer Jobs – Job (cycles/sec, Mbytes of memory) represents computational load for operations Messages – Message (MsgType, Size) represents communication load for communications
SOC-DEVS: Networked Interactions
9
SOC-DEVS: Simulation Example Real Voice Communication
System Streams End-to-End VoIP audio data to
subscribers Supports audio sampling rates and data
encryption 44.1 ~ 220.5 KHz 256 Key DES encoding 0% or 100% encryption
Supports multiple subscribers simultaneously System QoS is measured by the
VCS throughput Inter data frame delay
VCS Modeling in SOC-DEVS The real VCS is modeled
Models End-to-End VoIP audio data with sampling rates and data encryption 44.1 ~ 220.5 KHz 256 Key DES encoding 0% or 100% encryption
Simulation testbed is configured with similar configurations as in real VCS
Category Real System
Simulation System
Processor (CPU, Memory, Network Card)
2.2 GHz,1024 MB, 100 Mbps
2.2 GHz,1024MB,100Mbps
Network Link Bandwidth 100 Mbps 100Mbps
Subscriber # 1-40 1-40, 100-1000
Data Collection Duration
60 sec (wall clock)
60 sec (logical clock)
Tabl
e 1:
Sys
tem
con
figur
atio
n
Real System web services are developed in C# .NET
10
Testbed The testbed consists of
Real system Voice Communication System
Support up to 40 simultaneous clients
Automated data collection mechanism Throughput Delay
Packet level tracing Netmon 3.4
Simulation system Voice Communication System
Arbitrary VCS configuration Larger scale systems
DEVS-Suite simulator Transducer based data
collection Data analysis system
MATLAB scripts
Real System
Simulation System
Data AnalysisSystem
Testbed
Data
Analysis Output
Supports experimentation, data collection and data analysis
Round Trip Delay Definition
RT (Round Trip ) delay Client request sending event to first data arrival event Consists of
Server processing delay Network delay DelayServer processing + 2xDelayNetwork
Measured at client end ET2 – ET1
VoiceComm Network Client
delaynetworkdelayserver processing
1
2
ET = Event Time
Inter Frame Time
Inter Frame Time Time interval between
two consecutive audio frame events at the VoiceComm Service
Measured at server end IFTK = FTK+1 - FTK
K ={1,…N}
Frame1Frame 2Frame 3Frame 4VoiceComm
FT4 FT3 FT2 FT1
IFT2 IFT1IFT3
Accuracy
Accuracy The ratio of Total Bytes Received w.r.t. Total Bytes Sent
A = TBR / TBS Total Bytes Received (TBR)
Aggregated data bytes received by all the clients TBR = ∑ BR (K) ; K ={1,2,…N} and denotes Client ID
Total Bytes Sent (TBS) Aggregated data bytes sent by the VoiceComm service for all the
clients TBS = ∑ BS (K) K ={1,2,…N} and denotes Client ID
VoiceComm Network Client
Client
Delayserver processing Delaynetwork
Experiment Scenario
Client requests via network for audio data from the VoiceComm service
VCS sampling rate 44.1-220.5 KHz
VCS buffer size 16K
Client number 5-20
3 machines M1, M2, M5 connected via network M2 and M5 acts as clients using
multiple threads VoiceComm service sends data
for 60 seconds to each client Data is collected at probe points
Each configuration has 10 runs Data is averaged over these 10