Skip to content

Demand response programs are just hosts outsourcing their risk to you

Datacenter Talk by admin 25 replies 2.9K views
12 #1

Demand Response programs: a necessary evil or risk transfer?

We've seen increased discussion of DR enrollment across the industry. As a reminder, HostHatch participates in grid stabilization programs in applicable regions. This can mean brief, controlled load reductions during peak events.

Our stance: participation is disclosed in your agreement (Section 14.3), events are capped at 20 minutes, and credits offset any compute time lost.

However, we've noted some competitors are auto-enrolling customers without clear opt-out paths. We're reviewing our own communication clarity.

Questions welcome. — Admin

sudo make me a sandwich
#2

1. DR events violate uptime promises de facto, regardless of contract language.
2. "Controlled" is meaningless for:
- Database clusters with <30s failover
- Real-time inference pipelines
- Financial tick processing
3. Specific example: CloudCone DR event on 2026-05-14 lasted 23 minutes, not 20. Logs available.

Recommendation: negotiate explicit DR exclusion or demand 99.999% SLA with liquidated damages. Anything less is uncapped risk transfer.

It's always DNS. Always.
#3

23 minutes vs 20 cap—did CloudCone explain the overrun?

#4
magicmahdi said:
23 minutes vs 20 cap—did CloudCone explain the overrun?

They called it "grid operator directive beyond contractual limits." No credit offered beyond the standard minute-for-minute, which doesn't cover the cascade failure when my Postgres primary got terminated mid-checkpoint and the replica promotion timed out.

23 minutes is also their number. My external monitoring shows 26m14s from first connection refused to all-clear.

It's always DNS. Always.
9 #5

Same here. CloudCone 2026-05-14, monitoring from three locations. 24-27 minute window depending on probe. Their status page never went red, just "degraded performance." It was a hard power event, not throttling.

5 #6

I am in Casablanca. Our grid has instability every summer afternoon. If a European or American host wants to shed load, they do it during their peak, which is our evening. My users do not care about ISO New England's problems.

I would accept DR if I could pick the time window. I cannot. So I do not use hosts with DR in regions where my users are awake.

2 #7
FlowSana said:
Negotiate explicit DR exclusion

Good luck with that. I asked HostHatch sales directly last month. Their response: "DR participation is a facility-level agreement, not per-customer." The only opt-out is to not be in a facility that participates, which rules out most of their London inventory.

I stayed with them because Zurich was available and EKZ grid has different rules. Cost me 40% more.

#8
fiberwilliam said:
Zurich was available and EKZ grid has different rules

EKZ does run demand response, but it's frequency containment, not load shedding. Different mechanism. Your VPS stays up unless there's a genuine grid emergency, which last happened in 2021.

I track this because IPv4 isn't the only thing that only goes up. Swiss power reliability premiums do too.

/24 for sale. No lowballs.
#9

This is why I block the marketing emails and read the actual MSA. HostHatch's Section 14.3 is 847 words. CloudCone's DR clause is 94 words buried in an "Infrastructure Maintenance" section that doesn't mention demand response by name.

Both are disclosed. Neither is clear.

#10
admin said:
Events are capped at 20 minutes

My marriage lasted 23 minutes longer than that and I still consider it brief. "Capped at 20 minutes" is not a number you can build around. It's a number you can maybe survive once.

I run Proxmox at home for the stuff that actually matters. The VPS is for convenience. If it goes dark for 20 minutes, I shrug. But I am not running a business on it, and anyone who is should know the difference.

Post a reply

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

Post reply Preview Save draft