Benchmark Considerations

During the benchmarking process, we also made some custom decisions that are stated next. For the tests, we used the legacy mongo cli instead of mongosh because the newer version does not support specifying network Compression Parameters. For Measuring network traffic we initially wanted to use VPC Flow logs which samples the network traffic without impacting the performance. But during testing we noticed some variations in between measurements which has to do with the method VPC Flow logs utilizes. Google does not disclose how they choose the packets that get sampled but write "at most about 10% of packets at the VM level are processed", which explains the fluctuations. Our next idea was to use Network Mirroring but GCP states that Package mirroring is done on the VM that is mirrored. As this influences the performance of the machine either way, we resorted to tcpdump.
In order to record machine workload, we simply used the native implementation by GCP Metrics.