Client and administration console configuration

Server discovery

By default, JPPF clients are configured to automatically discover active servers on the network. This mechanism works in the same way as for the nodes, and uses the same configuration properties, except for the discovery timeout:

 # Enable or disable automatic discovery of JPPF drivers
 jppf.discovery.enabled = true
 
 # UDP multicast group to which drivers broadcast their connection parameters
 jppf.discovery.group = 230.0.0.1
 
 # UDP multicast port to which drivers broadcast their connection parameters
 jppf.discovery.port = 11111
 
 # IPv4 address inclusion patterns
 jppf.discovery.include.ipv4 = 
 
 # IPv4 address exclusion patterns
 jppf.discovery.exclude.ipv4 = 
 
 # IPv6 address inclusion patterns
 jppf.discovery.include.ipv6 = 
 
 # IPv6 address exclusion patterns
 jppf.discovery.exclude.ipv6 = 

A major difference is that, when discovery is enabled, the client does not stop attempting to find one or more servers. A client can also connect to multiple servers, and will effectively connect to every server it discovers on the network.

A client can also create mutliple connections to each discovered server, effectively creating a connection pool that can be used for concurrent job submissions. The size of the connection pools is configured with the following property:

 # connection pool size for each discovered server; defaults to 1 (single connection)
 jppf.pool.size = 5

Each server connection has an assigned name, following the pattern: “driver-<n>[-<p>]”, where n is a driver number, in order of discovery, and the optional p is the connection number, if the defined connection pool size is greater than 1.

For instance:

• if we defined jppf.pool.size = 1, the first discovered driver will have 1 connection named “driver-1”
• if we defined jppf.pool.size = 2, the first discovered driver will have 2 connections named “driver-1-1” and “driver-1-2”

The inclusion and exclusion pattern definitions work exactly in the same way as for the node configuration. Please refer to section Node configuration » Server discovery for more details.

Additionally, it is possible to specifiy the behavior to adopt, when a driver broadcasts its connection information for multiple network interfaces. In this case, the client may end up creating multiple connections to the same driver, but with different IP addresses. This default behavior can be disabled by setting the following property:

 # enable or disable multiple network interfaces for each driver
 jppf.pool.acceptMultipleInterfaces = false

This property is set to false by default, meaning that only the first discovered interface for a driver will be taken into account.

Manual network configuration

As we have seen, a JPPF client can connect to multiple drivers. The first step will this be to name these drivers:

 # space-separated list of drivers this client may connect to
 # defaults to “default-driver”
 jppf.drivers = driver-1 driver-2

Then for each driver, we will define the connection and behavior attributes, including:

Connection to the JPPF server

 # host name, or ip address, of the host the JPPF driver is running on
 driver-1.jppf.server.host = localhost
 # port number the server is listening to for connections
 driver-1.jppf.server.port = 11111

Here, driver-1.jppf.server.port must have the same value as the corresponding property jppf.server.port defined in the server configuration.

Backward compatibility with JPPF v2.x: To avoid too much disruption in applications configured for JPPF v2.x, JPPF will use the server port defined with the "old" property "driver-1.class.server.port" if "driver-1.jppf.server.port" is not defined.

Connection pool size

 # size of the pool of connections to this driver
 driver-1.jppf.pool.size = 5

This allows the creation of a connection pool with a specific size for each server we connect to, whereas all pools would have the same size when server discovery is enabled.

Priority

 # assigned driver priority
 driver-1.priority = 10

The priority assigned to a server connection enables the defintion of a fallback strategy for the client. In effect, the client will always use connections that have the highest priority. If the connection with the server is interrupted, then the client we use connections with the next highest priority in the remaining accessible server connection pools.

Connection to the management server

 # management host for this driver
 driver-1.jppf.management.host = localhost
 # management port for this driver
 driver-1.jppf.management.port = 11198

This will allow direct access to the driver's JMX server using the client APIs, unless the client configuration property jppf.management.enabled is set to false.

Using manual configuration and server discovery together

It is also possible to use the manual server configuration simultaneously with the server discovery, by adding a specific driver name, “jppf_discovery” to the list of manually configured drivers:

 # enable discovery
 jppf.discovery.enabled = true
 # specifiy both discovery and manually configured drivers
 jppf.drivers = jppf_discovery driver-1
 # host for this driver
 driver-1.jppf.server.host = my_host
 # port for this driver
 driver-1.jppf.server.port = 11111

Recovery and failover

As for the nodes, when the connection to a server is interrupted, the client will automatically attempt to reconnect to the same server. This is configured as follows, with the default values:

 # number of seconds before the first reconnection attempt
 reconnect.initial.delay = 1
 
 # time after which the system stops trying to reconnect, in seconds
 # a value of zero or less means it never stops
 reconnect.max.time = 60
 
 # time between two connection attempts, in seconds
 reconnect.interval = 1

With these values, we have configured the recovery mechanism such that it will attempt to reconnect to the server after a 1 second delay, for 60 seconds and with connection attemps at 1 second intervals.

Socket connections idle timeout

In some environments, a firewall may be configured to automatically close socket connections that have been idle for more than a specified time. This may lead to a situation where a server may be unaware that a node or client was disconnected, and cause one or more jobs to never return. To remedy to that situation, it is possible to configure an idle timeout on either side of the connection, so that the connection can be close cleanly and grid operations can continue unhindered. This is done via the following property:

 jppf.socket.max-idle = timeout_in_seconds

If the timeout value is less than 10 seconds, then it is considered as no timeout. The default value is -1.

Local and remote execution

It is possible for a client to execute jobs locally (i.e. in the client JVM) rather than by submitting them to a server. This feature allows taking advantage of muliple CPUs or cores on the client machine, while using the exact same APIs as for a distributed remote execution. I can also be used for local testing and debugging before performing the “real-life” execution of a job.

Local execution is disabled by default. To enable it, set the following configuration property:

 # enable local job execution; defaults to false
 jppf.local.execution.enabled = true

Local execution uses a pool of threads, whose size is configured as follows:

 # number of threads to use for local execution
 # the default value is the number of CPUs or cores available to the JVM
 jppf.local.execution.threads = 4

It is also possible to mix local and remote execution. This will happen whenever the client is connected to a server and has local execution enabled. In this case, the JPPF client uses an adaptive load-balancing algorithm to balance the workload between local execution and node-side execution.

Finally, the JPPF client also provides the ability to disable remote execution. This can be useful if you want to test the execution of jobs purely locally, even if the server discovery is enabled or the server connection properties would otherwise point to a live JPPF server. To achieve this, simply configure the following:

 # enable remote job execution; defaults to true
 jppf.remote.execution.enabled = false

Local execution flow

You can specify how frequently you wish to receive notfiications of locally executed tasks, using either or both of the following parameters:

 # specifies for how many completed tasks to wait until a notification is sent
 jppf.local.execution.accumulation.size = 4
 
 # specifies how long to wait before a notification is sent
 # (if any task has completed)
 jppf.local.execution.accumulation.time = 100
 # specifies the time unit:
 #   n = nanoseconds        M = minutes
 #   m = milliseconds       h = hours
 #   s = seconds            d = days
 jppf.local.execution.accumulation.unit = m

If both accumulation time and size are used at the same time, a notification will be sent whenever the size is reached, or the time is reached, whichever happens first. If neither is specified, the tasks wil be returned all at once.

Load-balancing in the client

The JPPF client allows load balancing between local and remote execution. The load balancing configuration is exactly the same as for the driver, which means it uses exactly the same configuration properties, algorithms, parameters, etc... Please refer to the driver load-balancing configuration section for the configuration details. The default configuration, if none is provided, is equivalent to the following:

 # name of the load balancing algorithm
 jppf.load.balancing.algorithm = proportional
 # name of the set of parameter values (aka profile) to use for the algorithm
 jppf.load.balancing.strategy = test
 # "proportional" profile
 strategy.test.performanceCacheSize = 2000
 strategy.test.proportionalityFactor = 1
 strategy.test.initialSize = 10
 strategy.test.initialMeanTime = 1e9

Also note that the load balancing is active even if only remote execution is available. This has an impact on how tasks within a job will be sent tot he server. For instance, if the “manual” algorithm is configured, with a size of 1, this means the tasks in a job will be sent one at a time.

Full client configuration file (default values)

 # list of drivers this client may connect to
 jppf.drivers = driver-1 driver-2
 
 # host name, or ip address, of the host the JPPF driver is running on
 driver-1.jppf.server.host = localhost
 # server port number
 driver-1.jppf.server.port = 11111
 # priority given to the driver connection
 driver-1.priority = 0
 # connection poool size for this driver
 driver-1.jppf.pool.size = 1
 
 # host name for the management server
 driver-1.jppf.management.host = localhost
 # port number for the management server
 driver-1.jppf.management.port = 11198
 
 # configuration for connection driver-2
 driver-2.jppf.server.host = my.host.com
 driver-2.jppf.server.port = 11121
 driver-2.jppf.management.port = 12003
 driver-2.priority = 10
 driver-2.jppf.pool.size = 1
 
 # enable/disable automatic discovery of JPPF drivers
 jppf.discovery.enabled = true
 # UDP multicast group to which drivers broadcast their connection parameters
 jppf.discovery.group = 230.0.0.1
 # UDP multicast port to which drivers broadcast their connection parameters
 jppf.discovery.port = 11111
 
 # automatic recovery: number of seconds before the first reconnection attempt
 reconnect.initial.delay = 1
 # time after which the system stops trying to reconnect, in seconds
 reconnect.max.time = 60
 # automatic recovery: time between two connection attempts, in seconds
 reconnect.interval = 1
 
 # enable/disable local job execution; defaults to false
 jppf.local.execution.enabled = false
 # number of threads to use for local execution
 jppf.local.execution.threads = 4
 # specifies for how many completed tasks to wait before a notification is sent
 jppf.local.execution.accumulation.size = 4
 # how long to wait before a notification is sent
 jppf.local.execution.accumulation.time = 100
 # the time unit
 jppf.local.execution.accumulation.unit = m
 
 # name of the load balancing algorithm
 jppf.load.balancing.algorithm = proportional
 # name of the set of parameter values (aka profile) to use for the algorithm
 jppf.load.balancing.strategy = test
 # "proportional" profile
 strategy.test.performanceCacheSize = 2000
 strategy.test.proportionalityFactor = 1
 strategy.test.initialSize = 10
 strategy.test.initialMeanTime = 1e9