|
|
|
|
|
|
by t-writescode
1988 days ago
|
|
|
This is the top response on your profile, so I’m commenting here. Atlassian is developed and run in Australia, right? Have you tried running your website and doing the things we mention here from a VPN that goes through the United States or Europe to experience our latency to your servers overall? I haven’t done any testing myself on this, but if you’re doing serial requests and each request is an https call back and forth to Australia, that’s easily 200ms every request, even if total server time is milliseconds. And even if you support parallel, if you limit the number of requests per non-whitelisted IP by a lot, it can very easily become approximately serial |
|
|
For the most part, each individual top line products' team operates out of a different site, with Confluence operating in Silicon Valley (mostly). I assume this is public information (or derivable) so I think it's safe to share.
Internally we have a few Confluence Cloud instances we use, with a single main one shared by the entire company (basically) which is located in a single area (somewhere?), though the VPN connection points are different.
So from a network topology standpoint our experience shouldn't be too different from most customers (at a very gross level) -> summary is, we definitely have users representative of 'bad networking', but you're right maybe I should be trying to intentionally degrade mine (right now i think its two cross US hops, but could be wrong).
I'll make sure our team looks at this dimension -> I don't think it's possible within our telemetry data, but maybe simulating it will get some interesting results.
I do recall reading once about something about AWS infrastructure (endpoints? edges?) having some configuration that causes omething like this: 1 - first returned packet is some some small size 2 - next packet (after ack is returned) can be double the size 3 - same 4 - same 5 - until max packet size
And though (1) is configurable (in theory), and I think the doubling/size increase is configurable (in theory), AWS does not allow for this configuration in their services.
But I can't find what I was reading so if anyone knows about that (a) being a problem and (b) how to work around it, let me know!