Hello,
Introduction
I've been running performance tests with 1 publisher and 1 subscriber with the following commands:
Publisher:
start cmd /k perftest_java.bat -pub -instances 10 -keyed -dataLen 40980 -domain 2 -latencyCount 1000 -numIter 1000000 -bestEffort -sendQueueSize 1Subscriber:
start cmd /k perftest_java.bat -sub -instances 10 -keyed -dataLen 40960 -domain 2 -bestEffort
Error/Issue:
When running these tests I write the test results to a .csv file after the test has completed and this feature has worked for previous types of tests. However, within this test these are the recorded latencies of 3 runs of the same test:
| One-Way Latency (us): | |
| 2931 | |
| 1004 | |
| 2479 | |
| 0 | |
| 0 | |
| 0 | |
| 0 | |
| 0 | |
| 0 | |
| One-Way Latency (us): |
| 2735 |
| 0 |
| 0 |
| 0 |
| 0 |
| 0 |
| 0 |
| One-Way Latency (us): |
| 3198 |
| 2680 |
| 2606 |
| 0 |
| 0 |
| 0 |
| 0 |
| 0 |
| 0 |
| 0 |
Question:
It would be great to understand what is happening here? Are my test scripts right? If so, why are the results interestingly missing?
Thank you
Blitz3r
Hello,
- I think the first question would be understanding if the output is correclty obtained. If you run perftest with no scripts, what is the output you obtain? I mean, the printed output, is that correct?
- How many latency samples are you getting in the test? What is the throughput in samples per second?
- Why are you measuring latency in a throughput test scenario? Are you aware you are sending at the maximum throughput and that queues are going to be full, so you are obtaining the worse case scenario Latency?
- Why are you setting the sendQueue to 1? Is that experimentally better for your tests? It might be that it is slowing down your system.
Hi Javier,
Thank you for your reply.
Question: If you run PerfTest with not scripts, what is the output you obtain?
Answer: The output is the same. However, where there are 0 values, the output is not shown on the command line. Here is a screenshot:
Question: How many latency samples are you getting in the test?
Answer: 1000
Question: What is the throughput in samples per second?
Answer: Averages: Run 1 = 0.9mbps, Run 2 = 2.7mbps, Run 3 = 2.9mbps
Question: Why are you measuring latency in a throughput test scenario?
Answer: I want to find the worst case latency when there is a load on the network.
Question: Why are you setting the send queue size to 1? Is that experimentally better for your tests?
How does it slow down the system? Initially, I thought that it should increase the load to ensure continuous traffic on the network so that the latency at higher throughput is measured.
Thanks,
Blitz3r