'Packet loss between two compute engines in India and USA on GCP?
It is normal to see packet loss between two compute engines in India and USA on GCP?
This was premium tier network bandwidth.
Can anything be done to improve the latency and reduce packet loss?
Solution 1:[1]
Your test does not indicate a problem.
Do not use traceroute to verify packet reliability. You can use traceroute to determine the paths (routes) from A to B usually. The ICMP protocol is best effort and routers are not required to respond. Therefore, packet loss at the ICMP level means nothing.
Traceroutes use ICMP and normal applications use TCP or UDP. Perform testing using the protocol your application requires.
Solution 2:[2]
A few different things:
Indicated packet loss for the "middle" hops is likely just ICMP rate limiting and not "real" packet loss. You shouldn't be particularly concerned about "middle" losses unless you also see loss reported for your last-hop (ie the destination).
Premium vs Standard Tier is just about VM to/from Internet communication. It does not come into play for VM to VM communication, even when those VMs are communicating via their external/public IPs.
For VM to VM communication within GCP it is a best practice to use internal/private IPs.
You don't say where specifically in India or the US these two VMs each happen to be, but you can go to https://datastudio.google.com/c/u/0/reporting/fc733b10-9744-4a72-a502-92290f608571/page/70YCB and see measured median inter-region latency to use as a sanity check against what you are measuring yourself.
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|---|
| Solution 1 | John Hanley |
| Solution 2 | Rick Jones |