Plug-in for Singleton Service in Clustered environment

1

• Plug-in for Singleton Service in Clustered
  environment
• and
• improving failure detection methodology
• Advisor By
• Dr. Chung-E-Wang Srinivasa c Kodali
• Department of computer
  science
• California state university,
  Sacramento

2

• AGENDA
• Importance of Singleton service in clustered
  environment.
• Implementation of singleton service.
• Need for singleton service as a plug-in
• Handling failure detection
• Improving failure detection methodologies
• Future Developments

3

• Importance of Singleton service in clustered
  environment
• What is singleton service?
• Service which manages all the singleton objects
  across the clustered application.
• What are singleton objects?
• Singleton is a pattern which makes sure that only
  one instance of that object is present at any
  point of time in the application.
• Need for singleton services
• At a time if only one user need to access the
  JDBC connection.
• Chat room application in the clustered
  environment where multiple users will be sending
  messages and broadcaster should send the messages
  to all the clients. Here broadcaster should be
  only one instance.

4

• Implementation of Singleton Service
• Implementation requires all the singleton objects
  to be accessible only as RMI objects.
• Instances of the service code will be present in
  all the JVM Instances but only one instance of
  RMI will be started.
• If the RMI service is started at one instance
  then update of JNDI registry should take place
  which tells where singleton service is running.
• JNDI will maintained to know which JVM is running
  the service and other JVMs running on the
  system.
• If another server instance need access to
  singleton object then the instance should make
  RMI call to the instance where the singleton
  service is running by looking into the JNDI.

5

• Implementation of Singleton Service
  contd
• Service contains a set of API methods which helps
  to start the RMI service, binding to the JNDI
  and unbinding the service.
• Some of the API methods that plug-in will contain
  are
• start() - will create RMI service on particular
  server.
• Bind() - Bind particular address to the JNDI
  directory.
• unBind() Un bind the address from the JNDI
  registry.
• getServerInstances() - returns a list of server
  instances running at that point.

6

• Need for developing a plug-in
• Open source servers doesnt have the singleton
  service implementation.
• Plug-in will reduces the coding time and
  maintenance capabilities of singleton services in
  clustered environment.
• Plug-in will be compatible with different kind of
  J2EE application servers like Tomcat, Jetty,
  glassfish etc.,

7

• Handling failure detection
• In clustered environment if the service running
  the JVM fails, all the singleton objects will not
  be accessible. It can become single-point
  failure.
• If JVM running the singleton service is down, how
  would the system know?
• A monitor in the cluster will be pinging the
  system which is running the singleton service at
  equal interval.
• What if the monitor pinging the system is down?
• A Highly available singleton service need to
  developed.
• How would other system know that the service is
  down
• If the monitor cant ping the system, monitor
  should be able to start the another singleton
  service automatically and update the JNDI
  directory.

8

• Improving failure detection methodologies
• Methodologies to make the singleton service
  highly available i.e failure detection
  algorithms
• Heart beat method similar to the Ping system
  which mentioned above.
• Async-BFD ( Asynchronous Bi-directional
  forwarding detection method) In this method any
  two nodes( can be random node, neighbor node)
  should be able to establish a communicate channel
  for a period of time.
• In the channel the nodes should send messages
  (ping) each other. If the ping node gets the
  reply back then other node alive.
• This method can eradicate any single point
  failure.

9

• Future Developments
• This methodology can be extended to cluster the
  instances like JMS etc.,
• Electing server to start the singleton service
  (failover detection) can be challenging task. Can
  use using complex distributed algorithms like
  leader- election algorithm.

10

• Questions
• ??