Title: MSc Project
1MSc Project
Grid Services Monitor Long Term Monitoring of
Grid Services Using Peer-to-Peer Techniques
- Yin Chen
- Supervised by Dr Stuart Anderson
- 2003
2Content
- ? Requirements
- ? Architecture
- ? Design Implementation
- ? Implementation Outcome
3Grid Services Monitor
4Requirements
- ? Monitoring grid services behaviours
- ? Providing long-term historical information
- ? Reporting to an end-user or a prediction model
- ? Helping make prediction of services behaviours
5Requirements
- ? The system must be fault tolerant
- ? Should allow nodes to join and leave
dynamically - ? Must be able to scale with the grid
- ? Monitoring data must be distributed
6Grid Services Monitor
7Architecture
- ? Peer-to-Peer architecture
- ? Why P2P ?
- - Decentralized fashion
- - Much scalability
- - Better fault tolerance
- ? Traditional P2P system designed for file
sharing and storing - ? This project focuses on decentralized data
querying and retrieval
8System Architecture
Monitor Sensor
Monitor Sensor
Data Store
Data Store
Local Collector
Local Collector
Listener A
Listener B
Collector
Data Store
Requester
9Topology
10Pull Model
- ? Data delivered by Pull
- ? Push model the listener sends out
notifications to other peers - ? Advantages
- - Less network traffic data deliver only when
necessary - - Has NO time synchronisation problem collect
data from resources at the same time - - The requester determines query conditions,
data type etc., making data operation easier
11Response Mode
12Query Planning
13Grid Services Monitor
14Design Implementation
- ? Monitor Sensor
- ? P2P Transport Mechanism
- ? Decentralized Data Storage and Querying
15Monitor Sensor
- ? Implement GT3 ServiceLifecycleMonitor
- Create when an instance of service is
going to be created - Destroy when the instance is going to be
destroyed - PreCall the service is going to be
invoked - PostCall service invocation has finished
- PreSerializationCall input parameters are
going to be desterilized - PostSerializationCall input parameters
have been desterilized - ? Problem can NOT report failure of the service
instance - ? Solution use timeout
16P2P Transport Mechanism
- ? Based on JXTA platform
- ? Queries are scoped to one peer group
- ? A requester multicasts a request message, all
listeners response by sending back requested data - ? Controlling the messages not to flood the
network - - Data delivered by Pull
- - Reduce the size of messages Local
collector precooked data - - Reduce the volume of messages
- ? Timeout to finish a data collection process
17P2P Transport Mechanism
ListenerB
Monitor Sensor
Output Pipe
Data Store
Input Pipe
ListenerC
Local Collector
Peer Group
Input Pipe
Output Pipe
ListenerA
Input Pipe
Output Pipe
Collector
Data Store
Requester
18Data Storage Querying
- ? Using XML file light weight and platform
independence - ? Decentralized manner
- Insert handled by Monitor Sensor of each
Listener - Retrieve by Local Collector of each
Listener - Data join by Collector of Requester
- ? Data operations
- Insert inserts a ChildNode at the root of
log file - Query XPath
- Join adds PeerID as a part of
primary key to avoid - primary key replication
problems
19Grid Services Monitor
20Generating Monitoring Data
Gt3 Server keeps tracing and monitoring
lt?xml version"1.0" encoding"ISO-8859-1" ?gt -
ltloggt  ltservice name"Weather Service" date"Aug
14, 2003" start"95543 AM BST" end"95559 AM
BST" succeed"true" /gt  ltservice name"Weather
Service" date"Aug 14, 2003" start"95543 AM
BST" end"95600 AM BST" succeed"true" /gt Â
lt/loggt
A Grid Service having a monitor sensor on it
Sample of log record
21Collecting Monitoring Data
- ListenerA
- RequesterAgt request
- Finish calculating data
- RequesterBgt request
- Finish calculating data
- ListenerB
- RequesterAgt request
- Finish calculating data
- RequesterBgt request
- Finish calculating data
- ListenerC
- RequesterAgt request
- Finish calculating data
- RequesterBgt request
- Finish calculating data
- RequesterB
- ListenerBgtI got it
- ListenerCgtI got it
- Finish collecting data
- Data collected from 2 peers
- RequesterA
- ListenerAgtI got it
- ListenerCgtI got it
- ListenerBgtI got it
- Finish collecting data
- Data collected from 3 peers
22Displaying Data In Graphical Layout
Number of responding peers
Service name
Key
Line of total running times
X Axis indicates the duration of the data
Y Axis indicates successful or total running times
Line of successful running times
Marks on the line indicate the values
23Displaying Data In Table Layout
Service name
Number of responding peers
The date of period
Success time
Total running time
24Comparing Different Services
25Grid Services Monitor