VNETPERF OUTPUT INERPRETETION
Please help me to understand why clockskew in vnetperf output is coming in negative between node 1 and node 2.
[dbadmin@verticadb-1 ~]$ /opt/vertica/bin/vnetperf --condense -hosts 10.65.34.102,10.65.34.104
The maximum recommended rtt latency is 2 milliseconds. The ideal rtt latency is 200 microseconds or less. It is recommended that clock skew be kept to under 1 second.
test | date | node | index | rtt latency (us) | clock skew (us)
latency | 2021-09-23_23:14:22,385 | 10.65.34.102 | 0 | 46 | 2
latency | 2021-09-23_23:14:22,385 | 10.65.34.104 | 1 | 53 | -1596
Whereas in other nodes/hosts its showing possitive
Output from node 3 and node 4
[dbadmin@verticadb-1 ~]$ /opt/vertica/bin/vnetperf --condense -hosts 10.65.34.105,10.65.34.106
The maximum recommended rtt latency is 2 milliseconds. The ideal rtt latency is 200 microseconds or less. It is recommended that clock skew be kept to under 1 second.
test | date | node | index | rtt latency (us) | clock skew (us)
latency | 2021-09-23_23:12:29,348 | 10.65.34.105 | 0 | 56 | 1835
latency | 2021-09-23_23:12:29,348 | 10.65.34.106 | 1 | 60 | 511
Answers
Clock skew between hosts is detected by taking the difference between the time of each node and the time at the start of the operation. For this reason, even the local node has clockskew. This skew (or "drift") should have a magnitude less than one second (1,000,000 us). Negative skew is just as bad as positive skew, just seen from the other side.
Interesting read: https://en.wikipedia.org/wiki/Clock_skew
Thanks @JimKnicely for the explanation so here we can just ignore the skew as it is below 1 second any other ways to check network related problems as we had a network outage earlier and since then we gettting delayed reports from server connections and [vertica] [vjdbc] [100161] connection close errors in dbvisualizer.
Make sure you are Using Connection Keep-Alive in DbVisualizer!