Contabo 100 Mbps plan feels slower than promised

Contabo's 100 Mbps pipe means nowt if their peering to your eyeball ISP is congested. I've seen Hetzner boxes push better real-world throughput on a 50 Mbps tier because their upstream doesn't oversubscribe transit.

Latency, bufferbloat, and whether you're hitting peak hours matter more than the headline number.

This is exactly why I run iperf3 to OVHcloud's looking glass before I ever trust a speedtest.net screenshot. Contabo could hand you a gigabit symmetric port and you'd still crawl if their upstream blend is light on routes to your last-mile carrier.

I've got a RackNerd VPS in their Chicago pop that benchmarks clean through to my Cleveland Comcast circuit. Same box, same test, pointed at a Vultr peer in New Jersey? Half the throughput. Same "100 Mbps" label on both ends.

The monthly cap argument is a separate beast. I've watched providers sell 10 TB plans on oversubscribed 1 Gbps links while others meter 2 TB across properly provisioned 10 Gbps. The quota tells you nothing about contention ratio or burst commitment.

If Contabo won't show you their AS adjacencies or let you test against their edge, you're buying a number on a webpage. I always ask for a trial IP to mtr against before I cut an invoice. Anything less is just hoping their marketing matches their transit bills.

Contabo's pipe is just a number on paper, same as my electric meter before the homelab happened

I ran iperf3 against their Phoenix pop last Tuesday, 100 Mbps tier clocked 23 Mbps to my eyeball ISP. Same window, my Vultr VPS at 50 Mbps pushed 47. The difference? Contabo routes me through three extra AS hops to hit anything off their own network. Vultr peers directly with my regional.

Raw megabits never mattered. What counts is where those bits actually travel.

My dumpster Juniper taught me that. You can have a fat pipe to nowhere, or a skinny one that actually delivers. Contabo's network map looks like a spiderweb designed by someone who hates spiders.

I keep a OVHcloud looking glass bookmarked for exactly this. Before any new provider touches my stack, they get the iperf treatment. No exceptions. Speedtest.net is for people who believe marketing departments.

Anyone else mapping traceroutes before they map budgets? Or am I the only one with a spreadsheet full of AS paths and regret?

https://bgp.tools

Contabo's 100 Mbps badge is like slapping a 1000W PSU label on a 65W brick. My OVHcloud Altra box pulls 12W at the wall and moves more bits-per-joule than that oversold pipe ever will. Bandwidth is just marketing fluff when the silicon underneath is wheezing.

Protocol throttling is the quiet killer. I have watched RackNerd shape rsync into oblivion while HTTP flies untouched, the congestion invisible until you measure what matters to your workflow.

Contabo's 100 Mbps tier carrying 23 Mbps on iperf3 is not ambiguity in naming. It is oversubscription dressed in polite language, the same way Vultr once sold me "unmetered" that evaporated at the third terabyte. The words become shields for the provider, puzzles for the user.

I do not draft posts about speedtests anymore. I draft posts about the gap between the interface and the wall socket, the two spaces between what is promised and what is metered.

🍜

Peering is a cop-out. My Hetzner 50 Mbps pipe in Denver hits 48 sustained because they actually provisioned the backhaul. Contabo's just cramming subscribers onto a 10G wave and praying.

Raw megabits are a vanity metric. My Hetzner NAT box streams smoother than mates on "faster" plans because their routing actually goes somewhere sensible.

Home gear matters, sure, but that lets Contabo off the hook. I want to know what their actual committed information rate is, not what they print on the brochure.

FTP throttling? That takes me back to my first colo in 2019, some outfit in Mesa that shall remain nameless. Their "unmetered" came with a side of 2 MB/s on port 21 because "legacy fairness." I was moving ISOs for a pfSense build and thought my switch died.

Contabo doesn't shape by protocol in my experience, they just... underdeliver universally. My iperf to their Phoenix pop clocked 67 down on a 100 contract. No QoS voodoo, no port-specific grief, just a pipe that wheezes after 7 PM like everything else on that ring.

The "sharing" angle someone mentioned is the real gotcha. OVHcloud spells out 15:1 on their cheap tier. Contabo? Crickets. I asked in a ticket, got back "best effort enterprise grade." Translation: we bought one 10G wave and called it a day.

Hetzner at least gives you a CIR you can hold them to. Learned that the hard way when I needed to push a backup set to RackNerd during monsoon season. Consistent beats optimistic every time when your lab's chewing through the wall.

Geography's half the puzzle, but Contabo still owes you a CIR that doesn't evaporate under load. My OVHcloud ARM rig in Singapore talks to a RackNerd VM in Amsterdam—~230ms—and the throughput stays predictable because both sides honor their floor. Contabo's "up to" language lets them stuff ten subscribers on one PON segment, all chasing the same photons. You want bits that survive dinner-hour congestion, not a geography lesson.