Greetings!

This is a great question, thank you for asking it here. There are several potential causes that I would expect to be potentially related. I’m going to list out the ones that I can think of.

1. Normal fluctuation in performance on the tested end-point. This to say, two tests to one destination could be wildly different. While continually reproducible results are unlikely related to the destination, logic would dictate that it is a possibility. Multiple tests merely reduce the probability.

2. Hardware. As much as we’d love to be consistent across the board on hardware, this is not entirely realistic. Good hardware that performs within acceptable parameters will generally not be pulled unless we’re reasonably certain that failure is near. New deployments will, of course, consist of our latest hardware builds. Because of this you can technically be provisioned on two somewhat different systems within the same datacenter.

3. Congestion. Could be at the hypervisor or the switch. There can be reasonable congestion that does not qualify as a problem, because it does not harm reasonable expectations of usage. It could, however, still be enough to cause a difference when compared with a server not experiencing the same.

4. Software. VPN is a common one here, more specifically OpenVPN. Encrypted traffic has overhead, and often requires tweaking for the system it sits on currently. Some CPUs may also expose certain flags that improve performance over others.

The only thing I feel rather certain about is that this is not a result I expect to see under any normal conditions:

Upload: 4.18 Mbit/s

Especially given the download rate on the same test, and that up/down are not treated differently, I suspect something external is at play there. As you said it’s not causing you trouble, but it really stood out to me so I wanted to bring it up.

Jarland