You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
For completeness we have provided all data available on the build farm; this includes data from Cyclone DDS, Connext DDS, and four variants of FastRTPS available on the buildfarm. The four available variants of Cyclone DDS include both dynamic and static memory variants along with two asynchronous and synchronous variants. Not all tests included all variants but we have included what data was available. These plots compare the four variants of FastRTPS were used during testing, dynamic C++ asynchronous (abbreviated DCA), C++ static (abbreviated CS), dynamic C++ synchronous (abbreviated DCS), and asynchronous C++ (abbreviate CA). One variant of Cyclone RMW was tested, which is the C++ static version. The Connext RMW is also included for reference.
Appendix B: Full Mininet Experiments Report
Comparing RMW Implementations Across Bandwidths and Packet Losses
Comparison is between the rmw implementations rmw_fastrtps_cpp async, rmw_fastrtps_cpp sync, and rmw_cyclonedds_cpp sync, and is
varied across bandwidth limits (1000, 300, and 54) and message type/rate
combinations (PointCloud512k@30 and Array1k@30).
Data was collected using the run_experiments.py script, and uses mininet
to simulate adverse network conditions, potentially varying the bandwidth limit, packet loss,
and/or packet delay for each process.
Each experimental run consists of two processes, one containing a publisher and the other
containing a subscription.
Experiments are run for 15 seconds, and various statistics are collected, but for this
comparison only "number of messages sent/received per second" are considered.
Comparisons with Bandwidth Limited to 54Mbps
Comparison Publishing Array1k@30
The specific details for this experiment are as follows:
Memory and CPU Consumption Averages for Mininet Experiments
Following from plot 2.2.4, this section presents a comprehensive look at the CPU
and memory consumption averaged across all Mininet experiments broken down first
by bandwidth, then by message size, and finally by loss and vendor. The results
remain fairly consistent with mixed results in terms of CPU consumption and a
generally smaller memory footprint for Cyclone DDS sync.
CPU and Memory Consumption for Bandwidth 54Mb, Message Size 1k Array
CPU and Memory Consumption for Bandwidth 54Mb, Message Size 512k
CPU and Memory Consumption for Bandwidth 300Mb, Message Size 1k Array
CPU and Memory Consumption for Bandwidth 300Mb, Message Size 512k
CPU and Memory Consumption for Bandwidth 1000Mb, Message Size 1k Array
CPU and Memory Consumption for Bandwidth 1000Mb, Message Size 512k