ICT ALMA Common Software

Space shortcuts

Page tree

Problem

BDNT configuration options

Solution


throttling(MBytesPerSec)

Default: 0.0. (no throttling)

Availability: SenderFlow; C++ API, CDB.

Description: Bulk data NT flow sender sends out data as fast as it goes - as infrastructure allows.. This can lead for example to exaggerated the network, what can result in a lots of lost data, and re-transitions, slowing down other network activity including sending from other flow senders... If we want to limit the data rate of sending we can do this by specifying throttling(MBytesPerSec). If 0.0 is specified, what is the default value, means no throttling, send as fast as possible. The trotting is applied just for sendData. With throttling we can achieve two things::

  • prevent exaggeration of the network infrastructure: switches/routers, receivers' NICs
  • better distribution of traffic in case of parallel arrays where data are sent from multiple computers


enableMulticast and multicastAddress

Default:

  • enableMulticast: true (multicast is used)
  • multicastAddress: 225.3.2.1

Availability: ReceiverFlow; C++ API, CDB.

Description: If enableMulticast is set to true (what is the default value) it is used multi-cast to send/receive the data, if the value is false the receiver for particular flow listens on unicast address. The multicast address can be set by specifying: multicastAddress. The default multi-cast address is 225.3.2.1. The value of multi-cast address is taken just just if multi-cast is enabled! It is not recommended to use multli-cast in combination with TCP.


unicastPort, baseUnicastPort and useIncrementUnicastPort

Default:

  • baseUnicastPort 48000
  • useIncrementUnicastPort true
  • unicastPort: 0

Availability:

  • baseUnicastPort useIncrementUnicastPort: ReceiverStream; C++ API, CDB.
  • unicastPort: ReceiverFlow; C++ API, CDB.

Description: The unicast port, in case if multicast is disabled, is used in the following way. If unicastPort is set to value different than 0 then the value is taken, otherwise if useIncrementUnicastPort is true then unicast port is calculated (incremented by one for each flow) from baseUnicastPort. If neither condition ( unicastPort is not set, and useIncrementUnicastPort is set to false) is met DDS choose the port. By default incremental ports are used.

In case if the port is set a new DEBUG message is logged: Stream#Flow: ABC#Flow123 going to listen on unicast address, port: 4800


participantPerStream

Default: false. (use global DDS participant)

Availability: SenderStream, ReceiverStream; C++ API, CDB.

Description: It regulates if a DDS participant is created for each stream, or a global one is used. Each instance of DDS participant means also more resources like memory and threads, but on the other side also better performance. So if there is a problem with number of threads/memory it should be used a global participant that is created with the creation of the fists stream. Default configuration is to use global (shared) participant, so no participant per stream, but rather one global for all the streams in a process.

A new log message appears if global(shared) participant is in use: Going to use global participant for stream: MyStream


(base)QosLibrary

Default: empty (BulkDataQoSLibrary QoS library will be used)

Availability: SenderStream, ReceiverStream SenderFlow ReceiverFlow; C++ API, CDB.

Description: By specifying (base)QosLibrary we can set which QoS library shall be used. The given QoS librarian has to be defined in: $ACSDATA/config/bulkDataNTDefaultQosProfiles.xml. If (base)QoS library is not specified, it is used default QoS library: BulkDataQoSLibrary.

Please see also: How to configure BDNT to use TCP ?


cbReceiveProcessTimeoutSec and cbReceiveAvgProcessTimeoutSec

Default:

  • cbReceiveProcessTimeoutSec 0.010 sec
  • cbReceiveAvgProcessTimeoutSec 0.050 sec

Availability:

  • cbReceiveProcessTimeoutSec: ReceiverFlow; C++ API, CDB.
  • cbReceiveAvgProcessTimeoutSec: ReceiverFlow; C++ API, CDB.

Description: With this options we set the upper limit (timeout) in sec for single frame (cbReceiveProcessTimeoutSec=) and average frame (cbReceiveAvgProcessTimeoutSec) processing time. The (single) processing time is measured by execution time of a single cbReceive callback method of the receiver. The average processing time is the average of processing time between cbStart and cbStop. The size of single frame correspond to 64000 bytes (~64kBytes), so it is measured processing time of a 64000 bytes. If either of limits are exceeded the BDNT logs an error message.


-- BogdanJeram - 2013-07-17