←back to thread

167 points voytec | 2 comments | | HN request time: 0.418s | source
Show context
niemandhier ◴[] No.42202012[source]
I have workloads on hetzner data centers in Finnland and Germany. Hetzner owns part of that cable, to connect those sites.

Tracerout gives me a path like this spine5.cloud2.fsn1.hetzner.com -> core22.fsn1.hetzner.com -> core5.fra.hetzner.com -> core53.sto.hetzner.com -> core31.hel1.hetzner.com

Which is worse than before, but still works for me.

replies(1): >>42202523 #
voytec ◴[] No.42202523[source]
My route from Poland to dedi box leased from Hetzner in Finland currently goes through AS1299 / twelve99.net (Warsaw->Talinn->Helsinki) peering, and there's a slight increase in latency since the cable cut. On a normal day, it's Poland -> Hetzner in Germany -> Hetzner in Finland. I'm guessing that the situation hurts companies like Hetzner financially.
replies(4): >>42202673 #>>42204657 #>>42204854 #>>42205919 #
1. input_sh ◴[] No.42204854[source]
I don't think a slight increase in latency can hurt them financially in any way. From my own measurements between Hetzner's datacenters, the latency increased from 25 ms to 40 ms.

Mildly amusing to say this had a direct, measurable impact on me? Absolutely. Something to lose sleep over? Maybe in some extremely niche situations.

replies(1): >>42205914 #
2. wongarsu ◴[] No.42205914[source]
I'd assume the financial impact is from increased costs for traffic. For Hetzner's customers traffic is somewhere between free and $1/TB, compared to AWS's $90/TB. Hence Hetzner hosts a lot of high traffic websites, and their margins on traffic have to be pretty slim. And with the cheap low-latency connection between the DCs some customers are bound to run a lot of traffic between servers in different Hetzner DCs.

Now I don't know what their peering agreements look like, but it seems like they normally route everything through Frankfurt's internet exchange and now have to take some different routes.