All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Why RackNerd has lower packet loss than ColoCrossing and EthernetServers ???
johnvmevpstorage
Member
I ran ping over period of many hours got these results:
RackNerd (Ashburn) packet loss: 0.017% over 50+ hours
ColoCrossing (NY) packet loss: 7.22% over 8 hours. I re-ran tests and got similar numbers again. I raised a ticket with them and they said it's all fine.
EthernetServers (Frankfurt) packet loss: 1.5% over 4 hours.
I know RackNerd uses ColoCrossing datacenter so how is it even possible for RackNerd to have lower packet loss than ColoCrossing??
I am paying the $11/year plan for RackNerd and $12 for ColoCrossing and $15 for EthernetServers. So why do I get lower packet loss on the cheaper plan than the more expensive plan?
Why is packet loss from my (UK) location to RackNerd US location (0.017%) lower than to Frankfurt (1.5%), even though Frankfurt is closer to the UK?
I am extremely confused and would appreciate if anyone could explain.
Comments
Different routing (transit) blend.
They have/use different upstream companies.
There are (usually) many providers in a data centre, a company can use any combination of them to connect to the internet.
7.22% seems high to me.. Personally, I open tickets when it is above 1%.
You've discovered the power of bandwidth doubling
What kind of provider are you talking about? Do you mean the network e.g. NTT or Level 3/Twelve99?
Exactly. There are many of them and they can be mixed together in different ways.
Can you post traceroutes for each? And yes you're comparing different networks, of course they perform differently.
Also, what are you pinging?
UPDATE: I ran mtr with -T option (TCP) and got 0.0% packet loss for colocrossing over 25,000 packets, whereas with UDP I got 6.1% packet loss over 25,000 packets so I'm inclined to think that twelve99 is randomly dropping UDP and ICMP packets while allowing TCP traffic.
EDIT: I am an idiot. My mtr for UDP only contains first 7 hops. So I actually don't know the real packet loss rate for UDP. Here are the UDP mtr results:
Start: 2024-03-21T19:32:00+0000
HOST: my PC Loss% Snt Last Avg Best Wrst StDev
1.|-- _gateway 90.0% 250 0.7 0.6 0.5 0.8 0.1
...[my ISP]...
8.|-- ??? 100.0 250 0.0 0.0 0.0 0.0 0.0
9.|-- ldn-b11-link.ip.twelve99.net 60.4% 250 5257. 63.7 2.9 5257. 528.9
10.|-- ldn-bb2-link.ip.twelve99.net 94.0% 250 2602. 196.6 3.6 2602. 669.6
11.|-- nyk-bb1-link.ip.twelve99.net 19.2% 250 74.6 74.0 72.7 79.7 0.8
12.|-- buf-b2-link.ip.twelve99.net 0.0% 250 92.6 92.5 91.5 93.6 0.3
13.|-- hostpapa-ic-xxxxxx.ip.twelve99-cust.net 49.2% 250 2847. 107.0 81.9 2847. 245.2
14.|-- ??? 100.0 250 0.0 0.0 0.0 0.0 0.0
15.|-- ??? 100.0 250 0.0 0.0 0.0 0.0 0.0
16.|-- ??? 100.0 250 0.0 0.0 0.0 0.0 0.0
17.|-- host.colocrossing.com 89.6% 250 93.2 95.4 92.0 125.0 7.7
This is still bad for me though because I want to use UDP for health checks.
The packet drops seem to be random though, not prolonged, so maybe if I send extra packets when there is a packet loss then it will make up for the lossy network.
What are you pinging?
If you ping a device with 100% load you will get losses even if the network is the best in the world.
I am pinging my VPS which is a newly installed Linux system, not running anything.
You need to use a 'known good' destination.
mtr for UDP shows 90% packet loss, holy cow:
Start: 2024-03-21T19:32:00+0000
HOST: my PC Loss% Snt Last Avg Best Wrst StDev
1.|-- _gateway 90.0% 250 0.7 0.6 0.5 0.8 0.1
...[my ISP]...
8.|-- ??? 100.0 250 0.0 0.0 0.0 0.0 0.0
9.|-- ldn-b11-link.ip.twelve99.net 60.4% 250 5257. 63.7 2.9 5257. 528.9
10.|-- ldn-bb2-link.ip.twelve99.net 94.0% 250 2602. 196.6 3.6 2602. 669.6
11.|-- nyk-bb1-link.ip.twelve99.net 19.2% 250 74.6 74.0 72.7 79.7 0.8
12.|-- buf-b2-link.ip.twelve99.net 0.0% 250 92.6 92.5 91.5 93.6 0.3
13.|-- hostpapa-ic-xxxxxx.ip.twelve99-cust.net 49.2% 250 2847. 107.0 81.9 2847. 245.2
14.|-- ??? 100.0 250 0.0 0.0 0.0 0.0 0.0
15.|-- ??? 100.0 250 0.0 0.0 0.0 0.0 0.0
16.|-- ??? 100.0 250 0.0 0.0 0.0 0.0 0.0
17.|-- host.colocrossing.com 89.6% 250 93.2 95.4 92.0 125.0 7.7
Possibly your ISP?
MTR is always needed both ways to prove anything. Could be an asymmetric route 'Fake packet loss' in the middle doesn't mean anything until it follows down.
MTR from your servers at CC, etc. back to your ISP for a better idea of where the issue lies.
But why? What I care about is having low packet loss from my PC to my VPS. Since I'm going to be sending packets from my PC to my VPS all the time.
I'd be interested in seeing MTR's for all three hosts with the MTR's run for the same duration, and captured at the same time.
Perhaps you can take a screenshot or use code tags, in order to maintain formatting that's a little easier to read.
The way I read it that you were pinging from one VPS to another..
Well, if it is from your home, you need to include a known-good address to determine if your home's internet connection is causing issues.
I think the mtr -u is not right because I used nc and manually sent UDP packets and saw only around 10% packet loss with my colocrossing VPS.
iperf confirmed that the real UDP packet loss for my colocrossing VPS was 12%:
I know my home internet is fine because I get near zero (0.017%) packet loss pinging my RackNerd VPS from my home PC.
UPDATE:
I just ran iperf for 10 seconds on my EthernetServer VPS and saw 0% UDP packet loss:
So maybe the packet loss is only ICMP or maybe it doesn't happen all the time.
Alright....
I'd also say that you need to test a known good location. Using one of the larger websites or test IP of a huge cloud provider, for example. Maybe the issue is the network you're testing from, or to, or both, but you won't know until you test a known good location.
Do you mean I should ping 8.8.8.8 from my home PC, or ping 8.8.8.8 from my VPSes?
Exactly. I'd test both ways. That way you can compare between them and try to see which server's network is acting up.
Maybe bbc.co.uk too. It isn't multi-casted like 8.8.8.8 is.
I'll run the ping and mtr again, while at the same time pinging other known-good IPs, will have the results by tomorrow.
Should I raise a ticket with EthernetServers if the ping packet loss is greater than 1%?
I don't know how much packet loss is considered normal/acceptable.
Are you referring to multihoming?
Welcome to the Internet
You mean the RackNerd network is multi-homed?
3 hours of pinging my $20 @ColoCrossing dedi in Buffalo, NY from Portland, OR:
Packets: Sent = 10979, Received = 10978, Lost = 1 (0% loss), Reply from x.x.x.x: Approximate round trip times in milli-seconds: Minimum = 82ms, Maximum = 325ms, Average = 88msOkay I finished pinging IPs overnight got these results:
8.8.8.8: 0.005% packet loss
1.1.1.1: 0% packet loss (WOW!)
BBC: 0.005% packet loss
I had my ColoCrossing VPS moved to a different node, ran iperf3 again now 0% UDP packet loss:
Also ran ping for 1800 seconds, 0% packet loss. So all issues seems to be resolved on that side. Props to ColoCrossing support for responding quickly and effectively resolving the issue.
EthernetServers Frankfurt looking glass: 1.33% packet loss
EthernetServers Frankfurt VPS: 1.25% packet loss
So it looks like it's not an issue with my EthernetServers VPS because I get the exact same amount of packet loss pinging their looking glass. So I think it's unlikely that I'll be able to get this issue resolved (if it is indeed an issue), but I'll raise a ticket anyway and see what they say about it.
EDIT: I just realized that I fucked up my mtr by running multiple mtrs at the same time, apparently they collide and make the result unreadable.
Fuck I'm an idiot.
I just ran mtr both directions from my EthernetServer VPS to my AWS UK EC2 instance and got 0% packet loss in both directions out of 1000 packets.
So, I'm not sure what's going on.
Seems like the packet loss only happens when I'm pinging the EthernetServers VPS from my home PC and not when I'm pinging it from my AWS instance.
I'll raise a ticket with EthernetServers and see what they have to say about it.