org.directwebremoting.impl
Class DefaultServerLoadMonitor

java.lang.Object
  extended by org.directwebremoting.impl.AbstractServerLoadMonitor
      extended by org.directwebremoting.impl.DefaultServerLoadMonitor
All Implemented Interfaces:
ServerLoadMonitor, UninitializingBean

public class DefaultServerLoadMonitor
extends AbstractServerLoadMonitor

A smart implementation of ServerLoadMonitor.

What a browser does:

       connected disconnected connected ...
      ____________          ____________
      |          |          |          |
      |          |          |          |
 _____|          |__________|          |______
       [---cT---] [---dT---] [---cT---] ...
 

Where cT is the connectedTime and dT is the disconnectedTime.

We impose some limits: a maximum number of simultaneously connected browsers maxWaitingThreads, and the maximum number of connections per second maxHitsPerSecond.

We attempt to keep the actual waitingThreads and hitsPerSecond within bounds by varying connectedTime and disconnectedTime.

The system is in one of 3 modes: USAGE_LOW, USAGE_HIGH and USAGE_DIGG. The boundary between USAGE_LOW and USAGE_HIGH is called threadOut. The boundary between USAGE_HIGH and USAGE_DIGG is called hitOut.

The system starts in USAGE_LOW mode. This mode uses constant values of connectedTime=60 secs and disconnectedTime=0 secs. We could use much bigger values for connectedTime (like infinite) however the servlet spec does not enable servlet engines to inform us if the browser goes away so we check by asking the browser to reconnect periodically.

In USAGE_LOW mode we measure the number of clients using the number of concurrently connected browsers (waitingThreads), when this goes above maxWaitingThreads we move into USAGE_HIGH mode.

On entering USAGE_HIGH mode, the settings (initially) change to connectedTime=49 secs and disconnectedTime=1 sec. As the load increases the connectedTime decreases linearly from 49 secs down to prevent the hits per second from going above maxHitsPerSecond. If the connectedTime goes below 1sec then the mode switches to USAGE_DIGG. If the connectedTime goes above 49 secs then mode switches to USAGE_LOW.

Note: there is some danger of an overlap where the system toggles between USAGE_HIGH and USAGE_LOW. We need some way to prevent this from happening.

On entering USAGE_DIGG mode, the connectedTime changes to 0 secs, and the disconnectedTime changes to 2 secs (to keep the round trip time at 2 secs). The disconnectedTime alters to prevent the hitsPerSecond from going above maxHitsPerSecond (In USAGE_HIGH mode the connectedTime was altered). When the disconnectedTime would go under 2 secs, we switch back to USAGE_HIGH mode.

Author:
Joe Walker [joe at getahead dot org]

Constructor Summary
DefaultServerLoadMonitor()
           
 
Method Summary
 long getConnectedTime()
          What's the longest time that we should wait before asking the client to reconnect?
 int getDisconnectedTime()
          Controller for poll times.
 void setMaxHitsPerSecond(int maxHitsPerSecond)
           
 void setMaxWaitingThreads(int maxWaitingThreads)
           
 boolean supportsStreaming()
          If the server is not going to be streaming then we need to tell browsers to just use XHR rather than anything fancier.
 void threadWaitEnding(WaitController controller)
          A thread has just ended a wait period.
 void threadWaitStarting(WaitController controller)
          A thread is about to begin a wait period.
 
Methods inherited from class org.directwebremoting.impl.AbstractServerLoadMonitor
contextDestroyed, servletDestroyed, shutdownRandomWaitControllers
 
Methods inherited from class java.lang.Object
equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Constructor Detail

DefaultServerLoadMonitor

public DefaultServerLoadMonitor()
Method Detail

supportsStreaming

public boolean supportsStreaming()
Description copied from interface: ServerLoadMonitor
If the server is not going to be streaming then we need to tell browsers to just use XHR rather than anything fancier.

Returns:
true if the server will be supporting streaming

getConnectedTime

public long getConnectedTime()
Description copied from interface: ServerLoadMonitor
What's the longest time that we should wait before asking the client to reconnect?

Returns:
The maximum client connected time

getDisconnectedTime

public int getDisconnectedTime()
Description copied from interface: ServerLoadMonitor
Controller for poll times.

TODO: We should probably get rid of this and leave it to BasePollHandler?

Returns:
How long should this client wait until it next polls?

threadWaitStarting

public void threadWaitStarting(WaitController controller)
Description copied from interface: ServerLoadMonitor
A thread is about to begin a wait period. This can be used by implementations to dynamically adjust the poll timings.

Specified by:
threadWaitStarting in interface ServerLoadMonitor
Overrides:
threadWaitStarting in class AbstractServerLoadMonitor
Parameters:
controller - An object that we can use to control the wait

threadWaitEnding

public void threadWaitEnding(WaitController controller)
Description copied from interface: ServerLoadMonitor
A thread has just ended a wait period. This can be used by implementations to dynamically adjust the poll timings.

Specified by:
threadWaitEnding in interface ServerLoadMonitor
Overrides:
threadWaitEnding in class AbstractServerLoadMonitor
Parameters:
controller - An object that we can use to control the wait

setMaxWaitingThreads

public void setMaxWaitingThreads(int maxWaitingThreads)
Parameters:
maxWaitingThreads - the maxWaitingThreads to set

setMaxHitsPerSecond

public void setMaxHitsPerSecond(int maxHitsPerSecond)
Parameters:
maxHitsPerSecond - the maxHitsPerSecond to set

Copyright 2008