8. Release Notes¶

8.1. Compatibility 3.0¶

8.1.1. Using Security¶

Governance and Permission files have been updated to be compatible with the latest release for RTI Connext DDS, and are compatible with RTI Connext DDS 5.2.7 and greater.

If you are compiling RTI Perftest against 5.2.5, you will need to get the certificates from the release/2.0 branch. You can do that by using the following git command from the top-level directory of your repository:

git checkout release/2.0 -- resource/secure

8.1.2. Compilation Restrictions¶

RTI Perftest is designed to compile and work against the RTI Connext DDS 5.2.x, 5.3.x and 6.0.0 releases.

Certain features, however, are not compatible with all the RTI Connext DDS versions, since the build scripts make use of certain specific parameters in rtiddsgen that might change or not be present between releases:

The --secure and --openssl-home parameters will not work for versions prior to RTI Connext DDS 5.2.5.
Java code generation against RTI Connext DDS 5.2.0.x will fail out-of-the-box. You can disable this by adding the --skip-java-build flag. See the Known Issues section for more information and alternatives.
C# code generation against RTI Connext DDS 5.2.0.x is not supported. You can disable this by adding the --skip-cs-build flag.

8.2. Release Notes 3.0¶

8.2.1. What’s New in 3.0¶

8.2.1.1. Ability to use your own type in RTI Perftest (#33)¶

RTI Perftest now supports the ability to use your own custom type. It is possible to measure the performance of your own type.

The Custom Types feature allows you to use your own customized types instead of the one provided by RTI Perftest. It is designed in such a way that the number of changes in the code and configuration files is minimal.

8.2.1.2. RTI Perftest thread priorities can be configured via command-line parameter (#65)¶

For the Traditional and Modern C++ API Implementations, a new parameter, -threadPriorities, has been added to RTI Perftest. This parameter allows you to set the priorities on the different threads created by RTI Connext DDS and by the application itself.

This parameter accepts either three numeric values (whichever numeric values you choose) representing the priority of each of the threads or, instead, three characters representing the priorities. These characters are h (high), n (normal) and l (low). These parameters can be used as follows:

-threadPriorities X:Y:Z

Where:

X is for the priority of the Main Thread that manages all the communication. X is also used for the Asynchronous Thread when using large data.
Y is the priority for all the receive threads. This value will be used for the Receive Thread created by RTI Connext DDS. If -useReadThread (use waitsets) is used, Y is for the thread in charge of receiving the data.
Z is the priority for the Event and DataBase Threads created at the RTI Connext DDS level.

This feature will only work for RTI Connext DDS Professional. To see what values should be used for the different threads see the following information in the RTI Connext DDS Core Libraries Platform Notes:

The “Thread-Priority Definitions for Linux Platforms” table
The “Thread-Priority Definitions for OS X Platforms” table
The “Thread-Priority Definitions for Windows Platforms” table

8.2.1.3. Raw Transport Support (#77)¶

RTI Perftest now supports raw transport communications. This allows the application to conduct performance tests skipping the DDS protocol. The purpose of this feature is to allow the calculation of protocol overhead and time differences.

To run a test with this feature, the -rawTransport command line option is required.

The Raw Transport feature is only aviable for C++ and supports two kinds of transport protocols, UDPv4 and Shared Memory.

The Raw Transport feature allows the following configurations:

Multicast (only for UDPv4)
One-to-many communication (Pub -> Sub)
Latency Test / Throughput Test
Scan

Some of the command-line parameters that exist for DDS are not supported if -rawTransport is used.

For the command -peer, the behavior has been modified. You can use -peer to set a peer address and a new optional ID:

Syntax: -peer <x.x.x.x>|<x.x.x.x:id>

If no ID is provided, it’s set as zero.

You can set multiple peers; the maximum value of accepted peers is RTIPERFTEST_MAX_PEERS, which corresponds to 1024.

Example:

perftest_cpp -pub -rawTransport -peer 127.0.0.1:5 -peer 127.0.0.1:6

A new command-line parameter, -noBlockingSockets, has been added:

This parameter changes the blocking behavior of send sockets to never block.
It is only available when -rawTransport is set with UDPv4 as the protocol.
This parameter can reduce the lost packets.
CHANGING THIS PARAMETER FROM THE DEFAULT CAN CAUSE SIGNIFICANT PERFORMANCE VARIATIONS.

8.2.1.4. Support for RTI Connext DDS Micro 3.0.0 (#78)¶

Starting with this release, RTI Perftest will have support for RTI Connext DDS Micro 3.0.0 and above.

By using the --micro and the --RTIMEHOME path command-line options at compile time, RTI Perftest will generate code for RTI Connext DDS Micro and try to compile using cmake (the path for which can also be configured by a command-line parameter in the build script). In this case, the RTI Perftest executable will be placed similarly to RTI Connext DDS Professional’s executable; however, it will be named perftest_cpp_micro.

Most RTI Perftest features are available when using RTI Connext Micro; however, some command-line parameters and options are available only for RTI Connext DDS Professional. More information about the supported parameters can be found in the Command-Line Parameters examples section.

8.2.1.5. Build HTML and PDF documentation (#94)¶

RTI Perftest’s build script for Linux now offers the option to generate the HTML and PDF documentation from the .rst files in srcDoc.

8.2.1.6. Allow 3 differents addresses for -multicastAddr feature (#97)¶

In previous versions of RTI Perftest, the -multicastAddr command-line parameter only supported a single address as input. This behavior has been improved. In addition to providing only one address, this parameter also allows you to provide three different addresses for each of the three topics used by RTI Perftest (Throughput, Latency, and Announcement).

Both IPv4 and IPv6 addresses are supported and can be set together on the same input command. All the input addresses must be in multicast range.

If you specify only one address, RTI Perftest will use that one and the two consecutive ones: for example, if you give 1.1.1.1, RTI Perftest will use 1.1.1.1 + 1.1.1.2 + 1.1.1.3. The higher values supported are 239.255.255.253 for IPv4 and FFFF:FFFF:FFFF:FFFF:FFFF:FFFF:FFFF:FFFC for IPv6.

8.2.1.7. Display in RTI Perftest’s subscriber side if the type expected is large data (#123)¶

RTI Perftest requires you to specify on the subscriber side the Data Length parameter if the data to be received is larger than the MAX_SYNCHRONOUS_SIZE constant. This parameter is used to change from the regular TestData_t type to TestDataLarge_t (used for large data). However, this was not displayed anywhere in the summary shown by the subscriber.

This issue has been fixed. Now the subscriber will show a short message stating that it is expecting the large data type.

8.2.1.8. Added –compiler and –linker command-line parameters to build.sh (#152)¶

When building in Unix, you can now use the –compiler and/or –linker command-line parameters to explicitly specify to the build.sh script the compiler/linker executables that will be used by rtiddsgen.

8.2.1.9. Ease the execution of RTI Perftest in VxWorks (#167)¶

In previous releases, it was not clear how to run RTI Perftest in VxWorks: each command-line parameter had to be appended to the argv array inside publisher_main and subscriber_main in perftest_publisher.cxx. This required recompiling each time the parameters changed.

This behavior has been simplified: in order to run in VxWorks, you can call the perftest_cpp_main function and receive a simple string containing all the command-line parameters.

8.2.1.10. Support RTI Perftest on Android platforms (#186)¶

Although the code for RTI Perftest is supposed to be platform-independent, it might not work out-of-the-box for mobile operating systems, since it expects to be used in a terminal.

Starting in version 3.0.0, RTI Perftest can also be compiled and used for Android platforms, using the basic graphical interface generated by rtiddsgen to print the output of the application.

8.2.1.11. Support RTI Connext DDS 6.0.0 FlatData and Zero-Copy features (#211)¶

RTI Connext DDS 6.0.0 introduces RTI FlatData language binding and Zero-Copy transfer mode over Shared Memory.

RTI FlatData reduces the number of copies made when sending a sample from a DataWriter to a DataReader from four to just two by building samples where the in-memory representation matches the wire representation.

Zero-Copy transfer mode accomplishes zero copies by using the shared memory (SHMEM) built-in transport to send 16-byte references to samples within a SHMEM segment owned by the DataWriter. This does not only reduces the latency but also makes the latency independent of the sample size.

Starting in version 3.0.0, RTI Perftest supports RTI FlatData language binding and Zero Copy transfer over Shared Memory.

This feature is not available when compiling for RTI Connext DDS Micro.

8.2.1.12. Increase send_socket_buffer_size for the UDPv4 transport¶

In order to achieve better performance with dealing with Large Data, the send_socket_buffer_size property has been modified from 500KB to 1MB in the QoS file.

8.2.2. What’s Fixed in 3.0¶

8.2.2.1. Migrate RTI Routing Service XML configuration to 6.0.0¶

The RTI Routing Service configuration file has been updated and is now supported by RTI Routing Service 6.0.0.

8.2.2.2. Remove duplicate code on RTIDDSImpl when the topic name is checked (#99)¶

Each time a DataReader or DataWriter was created, the topic name was compared with all the default topic names (Throughput, Latency, Announcement), in order to get the proper QoS Profile Name. This led to a lot of duplicated code on the createWriter and createReader functions.

This behavior has been fixed by creating a new function getQoSProfileName that accesses a new map, _qoSProfileNameMap, which contains the three topic names and their corresponding profile names.

8.2.2.3. Fix incorrect parsing of the -executionTime command-line parameter (#102)¶

In previous releases, for the Traditional and Modern C++ API implementations, the -executionTime <sec> command-line parameter would ignore any invalid value for the <sec> parameter without any notification to the user.

This behavior has been fixed and unified for all the API implementations, which now show an error when finding a wrong value for the <sec> option.

8.2.2.4. Ensure compatibility for the Traditional and Modern C++ Implementation (#114)¶

Some of the changes added for #55 broke compatibility when compiling certain platforms with no support for C++11. This issue has been fixed.

8.2.2.5. Wait for all perftest executions to finish before finalizing participants factory (#120)¶

In VxWorks kernel mode, static objects are shared across different runs of the same RTI Perftest libraries/executables, and changes in one run would cause changes in the other. When finalizing the Participant Factory after deleting the participant of an RTI Perftest execution, an error about outstanding participants in the domain was printed. This error occurred because the Participant Factory was shared accross runs in the same machine; therefore, participants from other executions prevented the factory from being properly finalized.

This issue has been fixed by checking that the factory is empty of participants before finalizing it.

8.2.2.6. Fix incorrect behavior for the -unbounded command-line option when not using large data (#125)¶

In the 2.4 release, a regression was introduced: the use of -unbounded caused a failure when using datasizes from 28 to 63000 bytes. This issue has been resolved.

8.2.2.7. Update maximum sample size accepted by RTI Perftest (#136)¶

The maximum size of a sample accepted by RTI Perftest has been updated to be compatible with RTI Connext DDS 6.0.0. This new value is 2147482620 bytes.

8.2.2.8. Add option to enable latency measurements in machines with low resolution clocks (#162)¶

If the machine where RTI Perftest is being executed has a low resolution clock, the regular logic might not report accurate latency numbers. Therefore, the application now implements a simple solution to get a rough estimate of the latency.

Before sending the first sample, RTI Perftest records the time; right after receiving the last pong, the time is recorded again. Under the assumption that the processing time is negligible, the average latency is calculated as half the time taken divided by the number of samples sent.

This calculation only makes sense if latencyCount = 1 (Latency Test), since it assumes that every single ping is answered.

8.2.2.9. Stop using alarm function to schedule functions since it is deprecated (#164)¶

When using -executionTime <seconds> parameter, internally, RTI Perftest was scheduling a function call by using it as a handler when an ALARM signal was received. This ALARM signal was set to be signaled in the amount of seconds specified by the executionTime parameter using the alarm() function available in Unix-like systems; however, this alarm function has been deprecated or is even missing in some of RTI’s supported platforms.

This issue has been fixed by using a thread that sleeps for the amount of seconds specified, after which the thread calls the desired function.

8.2.2.10. Remove the use of certain static variables that caused issues in VxWorks kernel mode (#166)¶

When running two or more instances of RTI Perftest within the same machine in VxWorks kernel mode, some parameters were shared between instances. This sharing happened because static variables are shared across different runs of the same RTI Perftest libraries/executables, and changes in one run would cause changes in the other. This issue has ben fixed.

8.2.2.11. Use Connext DDS implementation for the milliSleep method in C++ (#180)¶

The PerftestClock::milliSleep() method has been modified in the Traditional and Modern C++ implementations to always use the RTI Connext DDS sleep functionality. This change makes the sleep functionality independent of the operating system.

At the same time, the code has been improved to avoid overflowing the time of the sleeping period.

8.2.2.12. Fix Bottleneck due to low SHMEM QoS resources settings¶

The QoS setting dds.transport.shmem.builtin.received_message_count_max was set based on the OS default receive buffer size for SHMEM and the size of the payload sent on a sample.

The resulting allocated space was too small and therefore the throughput was being limited.

The dds.transport.shmem.builtin.received_message_count_max and dds.transport.shmem.builtin.receive_buffer_size QoS settings have been increased to avoid this bottleneck.

8.3. Release Notes 2.4¶

8.3.1. What’s New in 2.4¶

8.3.1.1. Summary of test parameters printed before RTI Perftest runs (#46)(#67)¶

RTI Perftest provides a great number of command-line parameters, plus the option of using the xml configuration file for modifying the RTI Connext DDS QoS. This could lead to some confusion with regards to the test that will run when executing the application.

In order to make this clear, RTI Perftest now shows a summary at the beginning of the test with most of the relevant parameters being used for thetest. The summary is done for both Publisher and Subscriber sides.

8.3.1.2. Added command-line parameters to simplify single API build (#50)¶

RTI Perftest Build scripts now support building a single API using the following command-line parameters:

–java-build –cpp03-build –cpp-build –cs-build

8.3.1.3. Added RTI Perftest and RTI Connext DDS information at beginning of test (#54)¶

RTI Perftest now prints at the beginning of the test its version and the version of RTI Connext DDS against which RTI Perftest is compiled.

8.3.1.4. Automatically regenerate qos_string.h file if perftest_qos_profiles.xml is modified (#63)¶

RTI Perftest now udpates the qos_string.h file with the content of perftest_qos_profiles.xml every time RTI Perftest is built for C++ and C++ New PSM.

8.3.1.5. Enable batching for Throughput-Test mode with 8kB value (#76)(#67)¶

As part of the enhanced out-of-the-box experience for RTI Perftest, batching is now enabled by default for throughput tests where the datalen is equal or smaller to 4kB. In such case, the Batch size value will be set to 8kB.

Batching will be disabled automatically if LatencyTest mode is set or if the -batchSize is lower than two times the -dataLen.

8.3.1.6. Use UDPv4 and Shared Memory as default transport configuration (#80)¶

Previously, the RTI Perftest default was to use only the UDPv4 transport. This did not, however, always lead to the best results when testing between applications within the same machine; it also differed from RTI Connext DDS default behavior, which enables the use of both UDPv4 and Shared Memory (SHMEM). Now, RTI Perftest’s new default behavior is the same as RTI Connext DDS: It enables the use of both UDPv4 and SHMEM.

This change improves the out-of-the-box user experience, getting better numbers when using the default configuration.

8.3.1.7. Show percentage of packets lost in subscriber side output (#81)¶

RTI Perftest now displays the percentage of lost packets in addition to the total number of packets lost. This percentage is displayed once per second with the rest of the statistics in the Subscriber side, as well as at the end of the test.

8.3.2. What’s Fixed in 2.4¶

8.3.2.1. Improved Dynamic Data Send() and Receive() operations (#55)¶

The Dynamic Data Send() and Received() functions have been optimized reducing the time spent setting and getting the samples.

As a result of these optimizations RTI Perftest now minimizes the time employed in application-related tasks, therefore maximizing the time spent sending and receiving calls. This allows a fair comparison between Dynamic Data results and Generated Type-Code Data results.

8.3.2.2. Corrected Latency maximum calculation in certain scenarios with low resolution clocks (#58)¶

In previous releases, if the clock provided by the system had low resolution, many of the Latency times calculated by sending and receiving back samples would end up being 0us. RTI Perftest would assume in those cases that this value was an initialization value and it would reset the maximum latency.

This problem has been fixed. RTI Perftest now correctly supports the case where the latency reported is 0us by not using it as a control/reset value.

8.3.2.3. Improved behavior when using the -scan command-line option and Best Effort (#59)¶

In previous releases, the use of -scan in combination with Best Effort would result in sending too many times certain packets used to signal the change of sizes and the initialization and finalization of the test.

In certain scenarios – mostly local tests where RTI Perftest Publishers and Subscribers were in the same machine and that machine had limitations with respect to the CPU – this problem would cause the Scan test to not work properly, since the Publisher would make use of the CPU and network intensively, potentially starving the Subscriber side and making the test hang.

This problem has been fixed.

8.3.2.4. Reduced memory consumption on Subscriber side (#74)¶

The initial_samples value for the ThroughputQoS QoS profile has been updated to a lower number. This profile is used by the Subscriber side to create a DDS DataWriter.

This value has been updated in order to decrease memory consumption on the RTI Perftest Subscriber side.

In order to ensure that this change does not affect the overall performance of the application, the initial burst of samples sent by the Publisher side has been also reviewed. The Publisher side now always send a burst big enough to ensure that the allocations in both Publisher and Subscriber sides are done before the test starts.

8.3.2.5. Fixed compilation in Certain VxWorks platforms (#93)¶

In previous releases the Traditional and Modern C++ implementations were tried to include sys/time.h, but this file might not exist in certain operating systems including certain VxWorks platforms.

This issue has been fixed, since this library is not needed in the VxWorks platforms, RTI Perftest excludes sys/time.h when compiling for VxWorks.

8.3.2.6. Migrate RTI Routing Service XML configuration to 6.0.0¶

The RTI Routing Service configuration file has been updated and it is now supported in the version 6.0.0.

8.3.2.7. Issues compiling in certain Platforms due to static variable transportConfigMap (#161)¶

In certain architectures the use of the static variable: static std::map<std::string, TransportConfig> transportConfigMap would cause some issues when referencing it from a static context.

In order to avoid this issue, the variable is not static anymore and it will be initialized in the constructor of the PerftestTransport class.

This issue affected both the Traditional and Modern C++ implementations.

8.4. Release Notes 2.3.2¶

8.4.1. What’s Fixed in 2.3.2¶

8.4.1.1. Traditional C++ Semaphore Take() and Give() operations not checking for errors properly (#47)¶

In previous versions, the semaphore Take() and Give() operations were not being checked for errors correctly in the Traditional C++ API implementation. This has been fixed.

8.4.1.2. Update Security Certificates and Governance files (#49)¶

The Security Certificates and Governance files used when enabling security options in RTI Perftest have been regenerated and signed again, since they had expired.

The script used for updating the files has been improved to generate certificates valid for a longer period of time (from one year to ten years).

8.5. Release Notes 2.3.1¶

8.5.1. What’s Fixed in 2.3.1¶

8.5.1.1. Keep Duration not configurable when using -noPositiveAcks (#39)¶

In previous versions, if the -noPositiveAcks command line parameter was provided the Disable Positive Acks Keep Duration QoS setting would be ignored both when provided via XML configuration or via command line parameter (deprecated option), instead, RTI Perftest would always use the default value set up via code.

This behavior has been fixed. We also took the oportunity to simplify and clarify the XML configurations when disabling positive Acks.

8.5.1.2. Show message in sumary when -multicast is present but it wont be used (#44)¶

In previous versions, if the -multicast command-line parameter was provided but the transport didn’t allow the use of multicast, it would fail silently and no indication would be shown by RTI Perftest.

Starting from this release, the use of multicast will be shown in the transport summary at the beginning of the test, and a message will be printed stating if multicast could not be applied for the transport.

The -multicast parameter has been divided into 2: -multicast which enables multicast for a given transport using a set of default multicast addresses and -multicastAddr <address> which enables multicast and sets the multicast IPs to be the one provided.

8.5.1.3. Update Security Certificates and Governance files (#49)¶

The Security Certificates and Governance files used when enabling security options in RTI Perftest have been regenerated and signed again, since they had expired.

The script used for updating the files has been improved to generate certificates valid for a longer period of time (from one year to ten years).

8.6. Release Notes 2.3.1¶

8.6.1. What’s Fixed in 2.3.1¶

8.6.1.1. Segmentation fault when using multiple publishers¶

In previous versions, in scenarios with multiple publishers, every RTI Perftest publisher application with -pidMultiPubTest different than 0 would crash in the process of printing the latency statistics. This behavior has been fixed.

8.7. Release Notes 2.3¶

8.7.1. What’s New in 2.3¶

8.7.1.1. Added Support for DTLS¶

RTI Perftest now supports the use of the DTLS plugin. The out of the box configuration allows the application to work using DTLS by just specifying -transport DTLS, however we also included command-line parameters to specify:

The Certificates and the public/private keys.
The verbosity.

See the Test Parameters section for more information about how to configure DTLS.

8.7.1.2. Added Support for TLS¶

RTI Perftest now supports the use of TLS on top of the TCP plugin. The out-of-the-box configuration allows the application to work using TLS by just specifying -transport TLS, however we also included command-line parameters to specify:

The Certificates and the public/private keys.
The verbosity.
The Server Bind Port.
The use of WAN mode.
The use of a Public Address.

See the Test Parameters section for more information about how to configure TLS.

8.7.1.3. Enhanced TCP Functionalities¶

As part of the changes for adding support for TLS, more functionalities have been included for TCP, including options to specify:

The verbosity.
The Server Bind Port.
The use of WAN mode.
The use of a Public Address.

See the Test Parameters section for more information about how to configure TCP.

8.7.1.4. Added Support for WAN¶

RTI Perftest now supports the use of the WAN transport plugin. In order to use this transport the command-line option -transport WAN needs to be specified, we also included command-line parameters to specify:

The WAN Server Address and Port
The WAN ID.
The Certificates and the public/private keys in case of using Secure WAN.
The verbosity.
The Server Bind Port.

See the Test Parameters section for more information about how to configure WAN.

8.7.1.5. Default Values for `Reliability` and `Transport` can be Modified via XML¶

Starting with this release, the Reliability and Transport settings are not set via code for the different languages, but are set in the XML profile. This allows you to easily modify these settings without needing to recompile.

These settings can still be modified via command-line parameters.

8.7.1.6. Added Command-Line Parameter `-qosLibrary`¶

Starting with this release, the QoS Library can be selected using the new -qosLibrary option.

This command-line option, combined with -qosFile, allows you to use custom QoS profiles that inherit from the default one (perftest_qos_profiles.xml).

A simple example is provided here: resource/profile_examples/custom_perftest_qos_profiles.xml.

8.7.1.7. Changed Name for Command-Line Option from `-qosProfile` to `-qosFile`¶

Starting with this release, the -qosProfile command-line parameter has been changed to -qosFile to better reflect its use.

8.7.1.8. Improved `-scan` Command-line Parameter Functionality¶

In the previous release, using -scan caused RTI Perftest to execute with a predefined set of values for -dataLen, and with execution durations related to the number of latency pings. This behavior has been changed. Now -scan allows you to specify a set of -datalen sizes to be used (or you can use the default set). In addition, the value specified for the ‘-executionTime’ parameter is now used for each execution during the scan, regardless of the number of latency pings.

When using -batchSize at the same time as -scan and not using large data, the same batch size will be applied to all the data sizes being used by -scan.

8.7.1.9. Deprecated Some Command-Line Parameters¶

To simplify the number of parameters RTI Perftest accepts, we reviewed and deprecated some parameters. These parameters will still work for this release, but they will be deleted or altered for future ones.

Deprecated -instanceHashBuckets <n>

The associated value will be the same as the number of instances.

Deprecated -keepDurationUsec <usec>

The value will be set in the QoS in the case of using -noPositiveAcks.

Combined -multicast and -multicastAddress <address>.

The resulting command can be used as -multicast keeping its original behavior or as -multicast <address>, which will enable multicast and use <address> as the multicast receive address.

Deprecated -nomulticast

The default behavior is to not use multicast, so this command-line option was redundant.

Updated -unbounded <managerMemory> to -unbounded <allocator_threshold>

Instead of managerMemory, use allocator_threshold, since it better reflects the use of the value. The new default is 2 * dataLen up to 63000. The associated documentation has also been improved.

Deprecated -heartbeatPeriod <sec>:<nanosec> and -fastHeartbeatPeriod <sec>:<nanosec>

These parameters can still be changed via XML.

Deprecated -spin <count>

This option made no sense after the -sleep and -pubRate alternatives were implemented.

8.7.2. What’s Fixed in 2.3¶

8.7.2.1. Failure when Using `-peer` Command-Line Parameter for C#¶

Using the -peer option in the C# implementation caused RTI Perftest to fail due to an issue reserving memory. This behavior has been fixed.

8.7.2.2. `-nic` Command-Line Parameter not Working when Using UDPv6 Transport¶

The -nic command-line parameter was not taken into account when using the UDPv6 transport. This behavior has been fixed.

8.7.2.3. Failure when Using -batchSize or -enableTurboMode if -dataLen Exceeded Async Publishing Threshold¶

Using -batchSize along with a -dataLen value greater than the asynchronous publishing threshold caused the application to show an error and exit. Starting with this release, the -batchSize option will be ignored in this scenario (and a warning message displayed).

This change (ignoring -batchSize) won’t be applied if you explicitly set -asynchronous; in this case, the behavior will remain the same as before (it will show an error and exit).

This change also applies to the use of -enableTurboMode.

8.7.2.4. Issues when Finishing Performance Test or Changing Sample Size¶

In order to make the mechanism to finish the performance test or change sample sizes more robust, we now use the Announcement topic on the Subscriber side to notify the Publisher side of the arrival of special samples sent to signal a change of sample size or to signal that the test is finishing. In previous releases, this process was not reliable and may have caused hangs in certain scenarios.

8.7.2.5. Unreliable Behavior Finishing Tests when Using ContentFilteredTopic (CFT)¶

In previous releases when using CFTs, in order to finish a test, the Publisher needed to send as many samples signaling that the test is finishing as the number of instances that were being used by the test (1 sample per instance). This could result in a very long process, and in scenarios where the reliability was set to BEST_EFFORT, in a higher chance of losing one of those samples, making the test hang.

This behavior has been modified by using a specific key for the signaling messages, so they are not filtered by the CFTs.

8.8. Release Notes v2.2¶

8.8.1. What’s New¶

8.8.1.1. Added command-line parameters “-asynchronous” and “-flowController `<`flow`>`”¶

In previous releases Asynchronous Publishing was only enabled for the DataWriters when the samples were greater than 63000 bytes and in such case, RTI Perftest would only use a custom flow controller defined for 1Gbps networks.

This behavior has been modified: Starting with this release, Asynchronous Publishing will be activated if the samples to send are bigger than 63000 bytes or if the -asynchronous command-line parameter is used. In that case, RTI Perftest will use the Default flow controller. However, now you can change this behavior by specifying the -flowController option, which allows you to specify if you want to use the default flow controller, a 1Gbps flow controller, or a 10Gbps one.

8.8.1.2. Improved “-pubRate” command-line parameter capabilities¶

In previous releases the “-pubRate” command-line option would only use the spin function to control the publication rate, which could have negative effects related with high CPU consumption for certain scenarios. Starting with this release, a new modifier has been added to this option so it is possible to use the both “spin” and “sleep” as a way to control the publication rate.

8.8.1.3. Added command-line parameter to get the CPU consumption of the process¶

Starting with this release, it is possible to display the CPU consumption of the RTI Perftest process by adding the Command-Line Parameter -cpu.

8.8.1.4. Better support for large data samples¶

Prior to this release, the maximum sample size allowed by RTI Perftest was set to 131072 bytes. The use of bigger sizes would imply changes in the perftest.idl file and source code files. Starting with this release, the maximum data length that RTI Perftest allows has increased to 2,147,483,135 bytes, which corresponds to 2 Gbytes - 512 bytes - 8 bytes, the maximum data length that RTI Connext DDS can send.

The sample size can be set via the -dataLen <bytes> command-line parameter. If this value is larger than 63,000 bytes RTI Perftest will enable the use of Asynchronous Publishing and Unbounded Sequences.

It is also possible to enable the use of Unbounded Sequences or Asynchronous Publishing independently of the sample size by specifying the command-line parameters unbounded <allocation_threshold> and -asynchronous.

8.8.1.5. Added command-line parameter “-peer” to specify the discovery peers¶

In previous releases the only way to provide the Initial Peers was either adding them to the QoS XML file or by using the environment variable NDDS_DISCOVERY_PEERS. Now it is possible to use a new command-line parameter: -peer <address> with the peer address.

8.8.1.6. Now providing RTI Routing Service configuration files to test performance along with RTI Perftest¶

A new configuration file and wrapper script have been added for testing RTI Perftest using one or several RTI Routing Service applications in between Publisher and Subscriber. A new section has been added to the documentation with all the configuration parameters: Using RTI Perftest with RTI Routing-Service.

8.8.1.7. Changed Announcement QoS profile to use “Transient local” Durability settings¶

In previous releases, the announcement topic DataWriters and DataReaders were set to have a Volatile Durability QoS. In certain complex scenarios, that could result in incorrect communication, which could cause the RTI Perftest Publisher and Subscribers to get stuck and not transmit data. By changing this topic to use Transient Local Durability, these scenarios are avoided.

This should not have any effect on the latency of throughput reported by RTI Perftest (as the main Throughput and Latency topics still have the same configuration).

8.8.1.8. Added new functionality: Content Filtered Topic.¶

In previous releases the only way to provide scalability was by using multicast and unicast. Now you can also choose which subscriber will receive the samples by using the parameter -cft. You can also determine which sample will be sent by the publisher with the parameter -writeInstance.

8.8.2. What’s Fixed¶

8.8.2.1. Conflicts when using “-multicast” and “-enableSharedMemory” at the same time¶

In previous releases, using “-multicast” in conjunction with “-enableSharedMemory” may have caused the middleware to fail while trying to access multicast resources although it was set to use only shared memory. This behavior has been fixed.

8.8.2.2. “-nic” command-line parameter not working when using TCP transport¶

In previous releases the -nic command-line parameter was not taken into account when using the TCP transport. This behavior has been fixed.

8.8.2.3. Batching disabled when sample size was greater than or equal to batch size¶

In previous releases the Batching Parameters were set unconditionally, now the Batching QoS will be only applied if the Batch size is strictly greater than the sample size.

8.8.2.4. Changed name of the “-enableTcp” option¶

In previous releases, the command-line option to use TCP for communication was named -enableTcpOnly. This is was inconsistent with other transport options, so the name of the command has been changed to -enableTcp.

8.8.2.5. Dynamic Data not working properly when using large samples¶

In previous releases the following error could happen when using the -dynamicData command-line parameter in conjunction with -dataLen greater than 63000 bytes:

DDS_DynamicDataStream_assert_array_or_seq_member:!sparsely stored member exceeds 65535 bytes
DDS_DynamicData_set_octet_array:field bin_data (id=0) not found
Failed to set uint8_t array

This error has been fixed starting in this release by resetting the members of the Dynamic Data object before repopulating it.

8.9. Release Notes v2.1¶

8.9.1. What’s New¶

8.9.1.1. Multicast Periodic Heartbeats when the `-multicast` command-line parameter is present¶

In previous releases, the Writer side sent heartbeats via unicast even if the command-line parameter -multicast was present. Now heartbeats will be sent via multicast when -multicast is used. This change should not affect one-to-one scenarios, but it will reduce the number of heartbeats the Publisher side has to send in scenarios with multiple subscribers.

8.9.1.2. Added command-line parameter to get the Pulled Sample Count in the Publisher side¶

The -writerStats command-line parameter now enables the some extra debug log messages shown in the Publisher side of RTI Perftest. These messages will contain the total number of samples being “pulled” by the Subscriber side.

8.9.1.3. Added extra logic to be able to support RTI Connext DDS 5.2.7 on Windows Systems¶

The names of the solutions generated by rtiddsgen for Windows architectures changed in Code Generator 3.2.6 (included with RTI Connext DDS 5.2.7). The solution name now includes the rtiddsgen version number. Therefore the RTIPerftest’s build.bat script now must query the rtiddsgen version and adjust the name of the generated solutions it needs to call to compile.

This change should not be noticed by the user, as the script will automatically handle the task of determining the version of rtiddsgen.

8.9.1.4. Added command-line parameter to avoid loading QoS from xml in C++.¶

If the -noXmlQos option is provided to RTI Perftest it will not try to load the QoS from the xml file, instead it will load the QoS from a string provided in the code. This string contains the same values the xml file provides.

This option is only present for the Modern and Traditional C++ PSM API code.

Note that changes in the xml will be ignored if this option is present.

8.9.1.5. Updated Secure Certificates, Governance and Permission Files¶

Governance and Permission files have been updated to be compatible with the latest release for RTI Connext DDS, and are compatible with RTI Connext DDS 5.2.7 and greater.

If you are compiling RTI Perftest against 5.2.5, you will need to get the certificates from the release/2.0 branch. You can do that by using the following git command from the top-level directory of your repository:

git checkout release/2.0 -- resource/secure

8.9.2. What’s Fixed¶

8.9.2.1. “–nddshome” Command-Line Option did not Work in `build.bat` Script – Windows Systems Only¶

There was an error in the build.sh script logic when checking for the existence of the compiler executable files. This problem has been resolved.

8.9.2.2. `build.sh` script did not make sure executable existed before starting compilation¶

Part of the build.sh script logic to check the existence of the compiler executable files was not being called properly. This issue is now fixed.

8.9.2.3. Incorrect `high_watermark` value when `sendQueueSize` is set to 1¶

Setting the command-line parameter -sendQueueSize to 1 caused RTI Perftest to fail, since it mistakenly set the high_watermark value equal to the low_watermark. This problem has been resolved. Now the high_watermark is always greater than the low_watermark.

8.9.2.4. Batching settings not correctly set in the `C++03` code¶

Settings related to batching in the XML configuration (perftest_qos_profiles.xml) were not being used. This problem has been resolved.

8.9.2.5. `dds.transport.shmem.builtin.received_message_count_max` incorrectly set in Java code¶

The dds.transport.shmem.builtin.received_message_count_max property was incorrectly set to 1 in every case. This erroneous behavior, which was introduced in RTI Perftest 2.0, has been resolved.

8.9.2.6. Command-line parameter for setting the RTI Connext DDS verbosity¶

In previous releases of RTI Perftest, the RTI Connext DDS verbosity could only be modified by using the command-line parameter -debug. This parameter would set the verbosity to STATUS_ALL, with no option to select an intermediate verbosity.

This behavior has been modified. The command-line parameter -debug has been changed to -verbosity, which can be followed by one of the verbosity levels (Silent, Error, Warning, or All).

The default verbosity is Error.

8.10. Release Notes v2.0¶

8.10.1. What’s New¶

8.10.1.1. Platform support and build system¶

RTI Perftest 2.0 makes use of the RTI Connext DDS rtiddsgen tool in order to generate part of its code and also the makefile/project files used to compile that code.

Therefore, all the already generated makefiles and Visual Studio solutions have been removed and now the build system depends on 2 scripts: build.sh for Unix-based systems and build.bat for Windows systems.

RTI Perftest scripts works for every platform for which rtiddsgen can generate an example, except for those in which rtiddsgen doesn’t generate regular makefiles or Visual Studio Solutions but specific project files. That is the case of Android platforms as well as the iOS ones.

Certain platforms will compile with the out of-the-box code and configurations, but further tuning could be needed in order to make the application run in the specific platform. The reason is usually the memory consumption of the application or the lack of support of the platform for certain features (like a file system).

8.10.1.2. Improved directory structure¶

RTI Perftest 2.0 directory structure has been cleaned up, having now a much more compact and consistent schema.

8.10.1.3. Github¶

RTI Perftest development has been moved to a GitHub project. This will allow more frequently updates and code contributions.

The URL of the project is the following: github.com/rticommunity/rtiperftest.

8.10.1.4. Numeration schema¶

RTI Perftest development and releases are now decoupled from RTI Connext DDS ones, therefore, and to avoid future numeration conflicts, RTI Perftest moved to a different numeration schema.

The compatibility between RTI Perftest versions and RTI Connext DDS ones will be clearly stated in the release notes of every RTI Perftest release, as well as in the top-level README.md file.

8.10.1.5. Documentation¶

Documentation is no longer provided as a PDF document, but as markdown files as well as in html format. You will be able to access to the documentation from the RTI Community page, as well as from the GitHub project.

8.10.1.6. Support for UDPv6¶

Added command-line parameter to force communication via UDPv6. By specifying -enableUdpv6 you will only communicate data by using the UDPv6 transport.

The use of this feature will imply setting the NDDS_DISCOVERY_PEERS environment variable to (at least) one valid IPv6 address.

8.10.1.7. Support for Dynamic data¶

Added command-line parameter to specify the use of the Dynamic Data API instead of the regular rtiddsgen generated code use.

8.10.1.8. Simplified execution in VxWorks kernel mode¶

The execution in VxWorks OS kernel mode has been simplified for the user. Now the user can make use of subscriber_main() and publisher_main() and modify its content with all the parameters required for the tests.

8.10.1.9. Decreased Memory Requirements for Latency Performance Test¶

The default number of iterations (samples sent by the performance test publisher side) when performing a latency test has been updated. Before, the default value was 100,000,000. This value was used to internally allocate certain buffers, which imposed large memory requirements. The new value is 10,000,000 (10 times less).

8.10.2. What’s Fixed¶

8.10.2.1. RTI Perftest behavior when using multiple publishers¶

The previous behavior specified that an RTI Perftest Subscriber in a scenario with multiple RTI Perftest Publishers would stop receiving samples and exit after receiving the last sample from the RTI Perftest Publisher with pid=0. This behavior could lead into an hang state if some RTI Perftest Publishers with different pid were still missing to send new samples.

The new behavior makes the RTI Perftest Subscriber wait until all the Perftest Publishers finish sending all their samples and then exit.

8.10.2.2. Possible `std::bad_alloc` and Segmentation Fault in Latency Test in case of insufficient memory¶

When performing a latency performance test with traditional or modern C++, the test tries to allocate certain arrays of unsigned longs. These arrays can be quite large. On certain embedded platforms, due to memory limitations, this caused a std::bad_alloc error that was not properly captured, and a segmentation fault. This problem has been resolved. Now the performance test will inform you of the memory allocation issue and exit properly.

8.10.2.3. Default Max Number of Instances on Subscriber Side Changed to `DDS_LENGTH_UNLIMITED`¶

In the previous release, if you did not set the maximum number of instances on the subscriber side, it would default to one instance. Therefore the samples for all instances except the first one were lost.

The new default maximum number of instances on the subscriber side has been changed from one to DDS_LENGTH_UNLIMITED. You can change this limit manually by setting the Parameter -instances <number>.

8.10.2.4. Error when using Shared Memory and Large Samples¶

When using RTI Perftest with large samples and enabling shared memory we could get into the following error:

Large data settings enabled (-dataLen > 63000).
[D0001|ENABLE]NDDS_Transport_Shmem_Property_verify:received_message_count_max < 1
[D0001|ENABLE]NDDS_Transport_Shmem_newI:Invalid transport properties.

8.11. Known Issues¶

8.11.1. Shared Memory issues when running the Modern C++ API or .Net Implementation¶

RTI Perftest uses UDPv4 and SHMEM by default; however certain operating systems don’t support Shared Memory, or the default configuration is not enough for RTI Connext DDS to work properly. In these cases RTI Perftest will show errors when trying to create the Participant entity:

[D0001|ENABLE]NDDS_Transport_Shmem_create_recvresource_rrEA:failed to initialize shared memory resource segment for key 0x40894a
[D0001|ENABLE]NDDS_Transport_Shmem_create_recvresource_rrEA:failed to initialize shared memory resource segment for key 0x40894c
[D0001|ENABLE]DDS_DomainParticipantPresentation_reserve_participant_index_entryports:!enable reserve participant index
[D0001|ENABLE]DDS_DomainParticipant_reserve_participant_index_entryports:Unusable shared memory transport. For a more in-depth explanation of the possible problem and solution, please visit http://community.rti.com/kb/osx510.
[D0001|ENABLE]DDS_DomainParticipant_enableI:Automatic participant index failed to initialize. PLEASE VERIFY CONSISTENT TRANSPORT / DISCOVERY CONFIGURATION.
[NOTE: If the participant is running on a machine where the network interfaces can change, you should manually set wire protocol's participant id]
DDSDomainParticipant_impl::createI:ERROR: Failed to auto-enable entity

These errors are handled and filtered in the RTI Perftest implementation for the Traditional C++ and Java APIs, but this is still not possible for the Modern C++ and .Net API.

For more information about how to configure Shared Memory, see http://community.rti.com/kb/osx510.

If you want to skip the use of Shared Memory in RTI Perftest, specify the transport using -transport <kind>, for example, -transport UDPv4.

8.11.2. Warning when compiling the Traditional C++ API Implementation¶

RTI Perftest might show these warnings when compiling the Traditional C++ API implementation for RTI Connext DDS Pro (in versions prior to 6.0.0) and for RTI Connext DDS Micro:

In file included from perftestSupport.h:15:0,
                from perftestSupport.cxx:11:
perftest.h:29:25: warning: ‘THROUGHPUT_TOPIC_NAME’ defined but not used [-Wunused-variable]
static const DDS_Char * THROUGHPUT_TOPIC_NAME= "Throughput";
                        ^
perftest.h:30:25: warning: ‘LATENCY_TOPIC_NAME’ defined but not used [-Wunused-variable]
static const DDS_Char * LATENCY_TOPIC_NAME= "Latency";
                        ^
perftest.h:31:25: warning: ‘ANNOUNCEMENT_TOPIC_NAME’ defined but not used [-Wunused-variable]
static const DDS_Char * ANNOUNCEMENT_TOPIC_NAME= "Announcement";
                        ^

These warnings are the result of a known issue in RTI Code Generator (rtiddsgen) (CODEGENII-873) related to the way in which the code for a const string is generated. This issue will be fixed in future releases of RTI Connext DDS Micro and has been already fixed for RTI Connext DDS Pro 6.0.0.

8.11.3. Building RTI Perftest Java API against RTI Connext DDS 5.2.0.x¶

Due to the changes added to support larger data sizes, RTI Perftest now makes use of Unbounded Sequences. This feature was not added to RTI Connext DDS in 5.2.0.x, so the following error will be reported when trying to compile the Java API:

[INFO]: Generating types and makefiles for java.
[INFO]: Command: "/home/test/nevada/bin/rtiddsgen" -language java -unboundedSupport -replace -package com.rti.perftest.gen -d "/home/test/test-antonio/srcJava" "/home/test/test-antonio/srcIdl/perftest.idl"
ERROR com.rti.ndds.nddsgen.Main Fail:  -unboundedSupport is only supported with C, C++, C++/CLI, or C# code generation
rtiddsgen version 2.3.0
Usage: rtiddsgen [-help]
. . .
INFO com.rti.ndds.nddsgen.Main Done (failures)
[ERROR]: Failure generating code for java.

To avoid this compilation error, two changes are needed:

In the build.sh or build.bat scripts, modify the call for rtiddsgen and remove the -unboundedSupport flag.
In the srcIdl/perftest.idl file, modify the TestDataLarge_t and TestDataLargeKeyed_t types, and add a bound to the bin_data member: sequence<octet,LIMIT> bin_data;.

8.11.4. Publication rate precision on Windows systems when using “sleep” instead of “spin”¶

When using the -pubRate <#>:sleep or -sleep command-line parameters on Windows systems, the sleep() precision will be accurate up to 10 milliseconds. This means that for publication rates of more than 10,000 samples per second we recommend using the “<#>:spin” option instead.

8.11.5. Compiling manually on Windows systems when using the RTI Security plugin¶

rtiddsgen-generated solutions for Windows systems allow four different configurations:

Debug
Debug DLL
Release
Release DLL

The new RTI Perftest build system, however, is focused on compiling only one of those modes at a time. To choose the compilation mode, use the -debug and -dynamic flags.

8.11.6. Warnings Compiling on Windows systems when using the RTI Security plugin¶

We have found that in certain instalations of Openssl a missing pdb file causes several warnings when compiling statically RTI Perftest for C++ (Traditional and Modern implementations). The warning that will show should be similar to this one:

libeay32z.lib(wp_block.obj) : warning LNK4099: PDB 'lib.pdb' was not found with
'libeay32z.lib(wp_block.obj)' or at 'rtiperftest\srcCpp03\objs\i86Win32VS2015\lib.pdb';
linking object as if no debug info [srcCpp03\perftest_publisher-i86Win32VS2015.vcxproj]

403 Warning(s)
0 Error(s)

This warning should be innocuous.

8.11.7. Dynamic compilation modes for RTI Connext DDS Micro¶

When building against the RTI Connext DDS Micro libraries, only the static compilation modes are supported. Therefore the --dynamic option will have no effect.

rtiddsgen code generator will fail with the following message: Option -sharedLib is not supported by this version of rtiddsgen.