Relatively high latency for short distance and odd routing for fibre to the home?

  • 20 November 2023
  • 3 reacties
  • 188 Bekeken

Reputatie 1

Hi all, 

I am hoping someone can shed some light on what is going on with the Odido peering and routing. In general I have been informed that fibre optic networks in have a 1ms latency per 100km. Now I am looking at my latencies, and I am used to seeing 2-5ms on a  100Mbps line at a distance of up to 1300km locally ( so not having to go through a sub-sea cable). Now locally for < 80km, I see 5-6ms? 
 

Also with the below picture why does < 500 packets show a different route being taking around hop2? I see this happen a lot, from my understanding a network should not be taking a different route so often.

 

Traceroute to Odido speedtest server in Amsterdam

 

Then if we look at https://www.submarinecablemap.com/ there are 8 cables directly between us and the UK, yet my latency is 12-16ms, and my packet has to first go to Sweden (I looked up the geo ip locations and owners of the IP address in the path)? That does not make any sense to me. The RENT Limited speed test server is ~380km from me, even if the fibre path was 500-600km it should still be half of what I am getting now.

Traceroute  to a RENT speed test server in the UK 

 

 

Now I am connected to my router with SFP+ DAC cables, my latency to my router is < 0.2ms, the above was run directly on the router. So if my PC has to run it, I would add 0.2ms to the timing, which can be ignored.  Just for reference: 

Traceroute from my PC, as you can see 0ms 

 

TechRacing93 5 maanden geleden

 Hi @ChaosMonkey

Allow me to try to answer some of your qeustions.

Also with the below picture why does < 500 packets show a different route being taking around hop2? I see this happen a lot, from my understanding a network should not be taking a different route so often.

 

Traceroute to Odido speedtest server in Amsterdam

 

 

Due to the probably redundancy and multiple link into the network, packets can be routed different based on various reasons. Looks like they are using ECMP from the looks of it, but that is just a guess. Btw, that is not the Odido speedtest server but Jonazz server, which is another ISP 😉. hence why you see hte 80.249.x.x address which belong to AMS-IX.

Then if we look at https://www.submarinecablemap.com/ there are 8 cables directly between us and the UK, yet my latency is 12-16ms, and my packet has to first go to Sweden (I looked up the geo ip locations and owners of the IP address in the path)? That does not make any sense to me. The RENT Limited speed test server is ~380km from me, even if the fibre path was 500-600km it should still be half of what I am getting now.

Traceroute  to a RENT speed test server in the UK 

 

First of all the GeoIP location does not always mean that the IP is exactly used by equipment located there. The 62.115.x.x are owned by Arelion Sweden. Which is as you can see here, one of the Transit partners used by Odido. 

Companies, especially Transit parties, set the GeoIP location to their HQ locations. With that said, It also depends which GeoIP service you ask. as you can see in the screenshot below one says Amsterdam, the other Stockholm 😉

 

I don’t see RETN in the list here. It seems there is no direct peering between Odido and RENT over for example the AMS-IX, hence why Arelion is used for transit. Looking at RETN website they are present in Amsterdam, looking at the avg response time I doubt the traffic leaves NL.

 

In general I have been informed that fibre optic networks in have a 1ms latency per 100km. Now I am looking at my latencies, and I am used to seeing 2-5ms on a  100Mbps line at a distance of up to 1300km locally ( so not having to go through a sub-sea cable). Now locally for < 80km, I see 5-6ms? 

Yess you are right, but that only applies for a direct point to point fiber connections. For example for host a to host b directly connected to each other. As that is not the case here the times you see are normal, taking all the routing into account. 

Im also a customer like you. It all be based on what i can find only and my own professional knowledge. 

 

Bekijk origineel

3 reacties

Reputatie 6
Badge +3

 Hi @ChaosMonkey

Allow me to try to answer some of your qeustions.

Also with the below picture why does < 500 packets show a different route being taking around hop2? I see this happen a lot, from my understanding a network should not be taking a different route so often.

 

Traceroute to Odido speedtest server in Amsterdam

 

 

Due to the probably redundancy and multiple link into the network, packets can be routed different based on various reasons. Looks like they are using ECMP from the looks of it, but that is just a guess. Btw, that is not the Odido speedtest server but Jonazz server, which is another ISP ;). hence why you see hte 80.249.x.x address which belong to AMS-IX.

Then if we look at https://www.submarinecablemap.com/ there are 8 cables directly between us and the UK, yet my latency is 12-16ms, and my packet has to first go to Sweden (I looked up the geo ip locations and owners of the IP address in the path)? That does not make any sense to me. The RENT Limited speed test server is ~380km from me, even if the fibre path was 500-600km it should still be half of what I am getting now.

Traceroute  to a RENT speed test server in the UK 

 

First of all the GeoIP location does not always mean that the IP is exactly used by equipment located there. The 62.115.x.x are owned by Arelion Sweden. Which is as you can see here, one of the Transit partners used by Odido. 

Companies, especially Transit parties, set the GeoIP location to their HQ locations. With that said, It also depends which GeoIP service you ask. as you can see in the screenshot below one says Amsterdam, the other Stockholm ;). 

 

I don’t see RETN in the list here. It seems there is no direct peering between Odido and RENT over for example the AMS-IX, hence why Arelion is used for transit. Looking at RETN website they are present in Amsterdam, looking at the avg response time I doubt the traffic leaves NL.

 

In general I have been informed that fibre optic networks in have a 1ms latency per 100km. Now I am looking at my latencies, and I am used to seeing 2-5ms on a  100Mbps line at a distance of up to 1300km locally ( so not having to go through a sub-sea cable). Now locally for < 80km, I see 5-6ms? 

Yess you are right, but that only applies for a direct point to point fiber connections. For example for host a to host b directly connected to each other. As that is not the case here the times you see are normal, taking all the routing into account. 

Im also a customer like you. It all be based on what i can find only and my own professional knowledge. 

 

Reputatie 1

The latency still seems a bit high even to AMS-IX at 7.8ms? I would have honestly expected that to be under 5ms considering a 1Gbps fibre connection. 
 

The only way a 12ms latency makes any sense is if it wasn’t in country. Even though it says RENT Amsterdam in the speedtest from ookla’s list, I would find it hard to believe that it takes 12ms to reach Amsterdam. The country is not that big. Unless the network setup is so bad here that it does indeed take 12ms for less than 80km distance? 

Thank you for pointing out the speedtest server was Jonas, I had taken it a while back. The odido AMS server gives me 4ms. But with regards to the multiple links, yes it would be setup for redundancy, but the route should not change that often. So I should not be seeing a routing change happen so soon and so often. This can cause unexpected latency issues in games because of that route change. There should be a weight attached to each route and it should keep to that unless a link becomes congested or goes down. Not a flip flop switch. 

From your https://bgp.he.net/AS50266#_peers link, RENT Limited is listed there, number 19 on that list? 

 

 

Reputatie 4
Badge

Might this point to a network that is on the verge of being congested, re-routing as soon as congestion is detected, and switching back when traffic eases (possibly through the re-routing)?
Congestion could also explain the spikes in ping time: waiting in a buffer takes a lot more time than travelling over a glass fiber.

Reageer