Closed ellerre closed 1 year ago
@ellerre,
Thanks for the report and sorry for the late reply. A few months ago, a look was taken to the throughput tests and some bugs were found. Some of them have been tackled, for instance #3309. Probably the weird results were related to this bug. I am going to proceed and close this issue but feel free to reopen if you get this weird results again.
Hello,
I am trying to benchmark FastDDS on fast Ethernet networks (1Gbps – 25Gbps – etc.). I chose to use the Throughput and Latency tests that you provide in the github repo (v2.3.0, commit bb500a5), but I have some problems in getting the expected throughput results for inter-host communication (latency is fine).
I've slightly modified the code of the test, since I want to have a single burst of exactly demand samples, sleep for recovery_time , and then go on with the next iteration (no timeout involved). In addition, I'm using RELIABLE reliability, KEEP_ALL history and VOLATILE durability.
I also want control on the network interface used for the test, and I want to use IP multicast for the test. Hence, I added the following lines to the publisher and the subscriber (I'm not using the xml file):
My testbed consists in two identical machines with:
I have increased the system-level UDP buffers as follows:
and tested with iperf that multicast works correctly in my testbed (>950 Mbps raw bandwith on server side). However, I am getting very strange results. Sometimes it seems like some of the sent messages are not actually received, even though my QoS settings require that, and in those cases performance is very low (or I get strange huge numbers). Plus, in general, I've noted that bandwidth is really volatile.
I have tried to use unicast instead of multicast. In that case, the behavior is much more stable, but performance is generally lower for big message sizes. Plus, I'm still getting those weird huge numbers every now and then. I've also tried to run the same test on much more modern and powerful machines with a 25Gbps network, but again I've obtained similar results.
Do you have any hint about what I’m doing wrong? I can provide additional information in case you need that. Thank you!