Skip to content

How do you test provider network without committing?

General Discussion by playersofia 6 replies 429 views
14 #1

I have been burned too many times by address space that looks abundant on paper but collapses under real routing conditions. My current threat model involves verifying whether a provider actually delivers clean legacy IP transit or if they are hiding carrier-grade NAT behind marketing slides. I want to measure latency, packet loss, and path diversity before I allocate any of my own address space to their machines.

What tools exist for this? I know of some public looking glasses, but most only show the upstream perspective, not the last mile. I am unwilling to deploy dual-stack as a compromise if the IPv6 side is an afterthought. NAT is not an option I will tolerate.

How do you validate network quality without first becoming a paying customer?

#2

Looking glasses are a start but honestly? The real test is finding someone with an existing machine and running iperf3 between your current box and theirs! Insane value when you can find a friendly user.

Ampere-based nodes usually have better NICs anyway, so if they run ARM you know the network stack is solid. Runs circles around x86 budget hosts on interrupt handling.

one small ping for man...
#3

I have had reasonable success with public traceroute servers. The key is to test from multiple vantage points, not merely one. Some providers operate their own looking glasses on their machines. Those are marginally more trustworthy than third-party tools.

One approach I do not recommend: exploiting trial periods to exhaustively benchmark, then cancelling. It is ethically dubious and wastes support resources.

-- Marcus

#4

Best practices for pre-purchase network validation:

  • Request MTR reports from sales engineering
  • Test via third-party looking glass services
  • Review published peering information at peeringdb
  • Verify SLA terms in service agreement

For prospective Hetzner customers, I asked for route optimization reports and latency matrices before I signed up.

I got the MTRs from their sales team and cross-checked them against a looking glass in Amsterdam. The numbers matched what they publish, so I went ahead.

#5

Iperf3 from a friendly user is the only way that actually works

rm -rf / --no-preserve-root ☯
#6

I don't trust provider looking glasses, they're just showing you the good paths

Vive la résistance... électrique
#7

What's an Ampere. Is that like the electricity.

phở at 3AM, deploy at 4

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft