Skip to content
Register

Tier III on the wall vs Tier III in reality

Datacenter Talk by FlowSana 24 replies 1.8K views
FlowSana

Member OP

SRE

3 #1
  • Uptime Institute's Tier Standard: Topology requires concurrent maintainability and fault tolerance for Tier III and IV respectively. Uptime Institute does not certify via self-assessment.
  • Common marketing gap: A facility may display a "Tier III design" certificate (dated, often 2019-era) while operating with single-path UPS configurations or non-concurrent maintenance windows in practice.
  • Specific example observed: A provider in the EU region advertised Tier III but ran single-module UPS with no maintenance bypass during Q1 2026. Their SLA was 99.982%, not the 99.982% expected from dual-path architecture.
  • Certification verification: Request the actual certificate number. Cross-check on uptimesite.com/certification (Uptime Institute's public registry). Verify scope: design, constructed facility, or operational sustainability.
It's always DNS. Always.
uma

Member

99.99% or bust

#2

I've got 47 hosts across nine "Tier III" facilities. Only three have never shown a maintenance window that took a path offline. The checklist I use for concurrent maintainability, stolen from a Contabo SRE off a conference hallway track — https://contabo.com —

  • Can they replace a UPS module without transfer to bypass?
  • Can they test a generator without switching to utility?
  • Can they maintain a CRAH while keeping N+1 cooling?
  • Do they publish MOPs 72 hours ahead?
Most "Tier III" shops fail on #1. The status page says "scheduled maintenance" and the graphs show a 40ms blip. That's not concurrent maintainability. That's a transfer event. My worst: a "Tier III" in Frankfurt where the ticket said "chiller maintenance" and my temperature alerts fired 90 minutes in. They had N cooling, not N+1. Design said N+1. Reality said otherwise. 99.982% on paper. 99.71% measured over 18 months.

436 days. reboot is surrender.
SP

Member

#3

Only three out of forty-seven. That tracks.

SP

Member

6 #4
uma said:
Can they replace a UPS module without transfer to bypass?

This is the one that gets me. I toured a facility in Virginia last year, big name, won't say who but they sponsor every KubeCon. Their "Tier III" setup had a single UPS string per feed. The sales engineer told me, straight-faced, that they "schedule maintenance during low-load periods" as if that changes the topology. Concurrent maintainability isn't about load. It's about architecture.

uma

Member

99.99% or bust

#5
SpockGone said:
As if that changes the topology

Exactly. I had that same conversation with a provider in Amsterdam. "We notify customers 14 days in advance." Great, so I can plan my outage? That's not what I pay for.

436 days. reboot is surrender.
norafilch

Member

#6

Hospital NOC here. We actually have Uptime Institute Tier III certification, the real thing, renewed 2024. The audit took three days. They traced every cable, verified every breaker interlock, watched us do a live generator test. The cost of that audit is why most colos skip it and just buy a "design" cert from 2019 and let it gather dust.

FlowSana

Member OP

SRE

3 #7
norafilch said:
The cost of that audit is why most colos skip it

This is the gap. A "Tier III design" certificate means an engineer once drew a diagram that looked compliant. A "Tier III constructed facility" certificate means someone verified it was built that way. The "Tier III operational sustainability" is the one that matters, and almost nobody has it.

It's always DNS. Always.
CR

Member

#8

So what's the move? Do you just assume everyone is lying and build for two regions?

uma

Member

99.99% or bust

#9
creekwire said:
Build for two regions?

I do. Active-active, split at the DNS layer. If a "Tier III" turns out to be Tier II with good marketing, traffic fails over. I've had two false alarms this year where my own health checks caught a blip the provider's status page never acknowledged.

436 days. reboot is surrender.
moebig

Member

RAID is not backup

#10

I run everything out of Casablanca. Local provider, no tier claims at all, just "we have two generators." I know exactly what I'm getting. The tier marketing is worse than useless, it creates a false floor.

RAID 1: because paranoia pays

Post a reply

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

Post reply Preview Save draft