Ten support tickets in ten years is what boring infrastructure looks like.
A national news publisher runs election nights on our hardware. In ten years it has opened roughly ten support tickets. This is what that took to build.
In this article 5 sections
- A national news publisher has been on our hardware for ten years and opened roughly ten support tickets in that time.
- Its election nights and championship finals run on our hardware. Nothing rings. That silence is the deliverable, not a happy accident.
- The stack is deliberately dull: clustered bare metal on HPE ProLiant, one known set of building blocks, engineers who answer your ticket.
- The same care scales down. One serious box runs a retail repair chain on exactly the same principle.
Good infrastructure has no story to tell. The metric nobody puts on a slide is the one that matters most: how often did anyone have to think about it? For the publisher the answer, across ten years of continuous high-traffic news, is roughly ten times. That is not a testimonial. It is a count.
What they needed
The brief was one sentence: when you need to reach your readers, there is no compromise with downtime. A publisher's peak is not a slow ramp you can watch on a dashboard and provision against. It is a verdict landing at 21:00, a goal in stoppage time, a resignation nobody saw coming.
What we built
We did not build anything exotic. We built one known thing, well, on dedicated hardware. More than 30 HPE ProLiant servers running clustered bare-metal workloads: HAProxy and Varnish at the edge, nginx and PHP, Percona XtraDB clusters, clustered Valkey and Elasticsearch, on HPE 3PAR all-flash storage with 3PAR Peer Persistence, with geographically distributed disaster recovery.
- COMPUTE
- HPE ProLiant, bare metal, clustered
- DATABASE
- Percona XtraDB Cluster (multiple 3-node)
- SEARCH / CACHE
- Elasticsearch · Valkey · Varnish
- STORAGE
- HPE 3PAR all-flash, Peer Persistence
- DISASTER RECOVERY
- Geographically distributed disaster recovery
The most telling part of the build is one its readers never saw. The platform moved from Amsterdam to Digital Realty Athens months ago, and the cutover was uneventful for its readers.
What changed
Now the election nights and the championship finals run, and support hears nothing.
Notice what did not happen. We did not sell the publisher a bigger meter or a premium support tier to survive their peaks. We sized the machine to the real peak and put the headroom in the box.
The receipt
You can argue with adjectives. You cannot argue with a ticket count. Here is the publisher's, against a decade of the most unforgiving traffic pattern in publishing.
The publisher's election nights happen. Nobody calls. Silence, in this line of work, is the entire product. Everything else is packaging.
Why this generalizes
None of what made the publisher's platform boring is specific to publishing. Owned hardware with no meter on it, one known stack instead of a zoo of half-understood services, and engineers who answer your ticket with a name instead of a queue. Those are the ingredients, and they do not require a national readership to work. Our customer stories show more of how we work.
The same care scales all the way down to one serious box. A retail repair chain runs its stores, its technicians, and its business-critical ERP on a single dedicated server. Different size, identical outcome: infrastructure that does its job and then gets out of the way.
Straight answers
How do I know this applies to my size?+
Because the principle is not about size. It is about owning the hardware, standardizing the stack, and having a named engineer on your ticket. That works at one dedicated box and at a full private cluster. The Sizing Desk will tell you which end of that range you actually need, and it errs toward the smaller answer.
What does "managed" include?+
Hands-on migration, running the platform day to day, and performance and security optimization, all on our side of the line. We build it, we run it, we answer when something needs attention. For the publisher that has meant roughly ten occasions in ten years.
What happens on my peak night?+
You burst into headroom that is already in the box, because it is dedicated hardware and there is no meter on it. We size that headroom to your real peak up front.
What happens if a whole datacenter has a bad day?+
The publisher's platform has geographically distributed disaster recovery, so a problem at one site is not a problem for its readers.
Size the cluster behind an uneventful election night
Answer a handful of questions and the Sizing Desk proposes a real architecture from our real building blocks. It will tell you if you need less than you think. Most people do.
Size your clusterYou’ve reached the end.
Put the reading to work
What are
you running?
Tell us about the workload, traffic and recovery requirements. Start with the server builder, or talk through the details with an engineer.
Written by the engineers who build and run the infrastructure it describes.Talk to an engineer