01 / Why it mattersThe difference is almost never the kit
A hall is commissioned once and operated for a decade. Redundant power, N+1 cooling and a second carrier give it the capacity to survive a fault. They do not decide whether it does.
That is settled later: by whoever is on shift, what they are allowed to do, how quickly they can reach the rack, and whether the procedure they follow was ever tested. Two identical halls, built to the same drawing, post very different availability figures after three years, and the difference is almost never the hardware.
This guide is about buying that operation rather than running it yourself: what the scope covers, how to read a response SLA, the tier model the market has settled on, and the arithmetic that decides which response time your uptime target can support. For the engineering discipline underneath it, from the availability stack and MTBF and MTTR to change control and the full incident lifecycle, read running a data centre: operations, resilience and uptime.
02 / ScopeWhat “operations” actually covers
Operations is often bought as one line item and then found to be six. Be explicit about scope before comparing anyone’s price, because the cheapest quote is almost always the one that covers the least.
This is the distinction that sets the price, and it is why comparing the two on a rate card gives the wrong answer. Smart hands is one of the six, bought by the visit. Managed operations is all six, held between visits as well as during them, which is what stops the other five quietly degrading.
- 01
Hardware & rack
Installs, moves, adds and changes, goods receipt and RMA.
- 02
Connectivity & cabling
Patching, cross-connects, labelling and record accuracy.
- 03
Capacity & assets
Power, cooling, space and U tracked against what is installed.
- 04
Monitoring & alerting
Telemetry with thresholds that somebody owns and acts on.
- 05
Security & access
Who may enter, with what authority and what audit trail.
- 06
Incident ownership
Someone named who owns the fault through to closure.
03 / Self-assessmentScore your operational maturity
Operational maturity is the gap between a site that has good kit and one that is run well. Score each discipline honestly: anything you cannot evidence with a document or a record is a gap, not a strength. The disciplines are weighted by how often they decide the outcome of an incident, so monitoring and procedure count for more than training does.
Pick the level you could evidence today
Nine disciplines, four levels each. Your weighted score and the three gaps that cost the most appear once all nine are scored.
Nothing is sent anywhere. The scoring runs in your browser and nothing is stored.
Monitoring and procedures are weighted highest deliberately: you cannot respond to what you cannot see, and you cannot act safely without a method. The most common failure pattern is not missing redundancy. It is redundancy that was never tested, or a procedure that existed only in one engineer’s head.
04 / IncidentsWhat happens when it goes wrong
Every discipline above is really preparation for a few minutes of someone else’s bad morning. This is the shape of those minutes, and the step a response SLA actually measures is the third one.
- 01
Detect
Automatically and within seconds, not by a user reporting it. Detection you do not own caps everything after it.
- 02
Alert & triage
Routed to someone awake, qualified and authorised, with severity set against a definition rather than a feeling.
- 03
Respond
To procedure, with the access, tooling and spares already arranged.
What the SLA measures - 04
Stabilise
Stop it getting worse before making it better. Restoring service and fixing the root cause are different jobs.
- 05
Restore & verify
Service back and proven back, measured rather than assumed, with redundancy restored before closure.
- 06
Review
Blameless, documented and turned into a change. An incident that alters no procedure will happen again.
05 / Service levelsHow to read an operations SLA
Response times are the most quoted and least examined number in this market. Four checks separate a commitment that will hold from one that only reads well in a proposal.
Time to hands-on, not time to fixed
A response target is the clock until someone qualified is at the rack. Diagnosis and repair come after it, out of the same outage budget. A response SLA presented as a fix SLA is the most common overclaim in operations contracts.
Where the spares are
Fifteen minutes to a rack you cannot fix is fifteen minutes. A fast response is only worth what the on-site spares make it worth, so ask what is held, where, and who funds it. If the answer is a next-day courier, the real response is next day.
How many people are on shift
One technician on site means the second concurrent incident breaches by definition, and lone working is not permitted on live electrical or liquid-cooled plant anyway. Genuine round-the-clock presence is at least two per shift: roughly eleven full-time staff once cover, leave and shift leads are counted.
What every tier commits to
Detection and notification should not vary with what you pay. A P1 detected, notified and ticketed inside fifteen minutes is reasonable to expect at the lowest tier as well as the highest. The tier should change what happens next, not how quickly you find out.
Ask a prospective provider what happens at 02:00 on a bank holiday when two racks fail at once. The answer exposes the shift pattern, the spares, the escalation path and the honesty of the response figure in a single question.
06 / The modelThe Optronix tier model: On-Call to Embedded
Optronix delivers managed operations on a five-step ladder. What separates the steps is not response time alone. It is presence, how often someone is physically on your floor, and depth, how much of the stack we own. Response time follows from those two, which is why buying a fast response without the presence to support it does not work, from us or from anyone else.
-
On-Call
No standing presence. We attend on request, with your runbook and asset register already held and maintained between visits.
- On site
- On request
- Response
- 48 hr / NBD
-
Scheduled
Fixed attendance days per site, working a managed ticket queue, with reactive cover in between.
- On site
- Set days a week
- Response
- 48 hr
-
Managed
Two or more days a week on site plus round-the-clock monitoring, with Optronix owning each incident through to closure.
- On site
- 2+ days a week
- Response
- 8 hr
-
Resident
Dedicated Optronix technicians on your floor every day across an extended two-shift pattern, on call overnight.
- On site
- Daily, two shifts
- Response
- 30 min
-
Embedded
A dedicated crew on site around the clock, spares held and a committed availability target. We own the operations function end to end.
- On site
- 24/7/365
- Response
- 15 min
At one end this is light touch: nobody standing on your floor, a runbook and an asset register kept current, and an engineer dispatched when you need one. At the other it is a fully outsourced operations team: technicians on site 24/7/365, owning the estate end to end, with spares on site and a committed availability target. Same model, same portal, same reporting. The only variable is how much of the operation you want to keep yourself.
Tiers 01 to 03 are priced on volume; 04 and 05 are priced on a rota, because they are staffing commitments rather than task commitments. That is also why the top two normally carry a minimum term and a mobilisation fee: nobody hires eleven people against a rolling monthly contract.
What sits at each step
| Capability | 01On-Call | 02Scheduled | 03Managed | 04Resident | 05Embedded |
|---|---|---|---|---|---|
| Installs, moves, adds and changes | |||||
| Ticketing portal, managed queue and service reporting | |||||
| Goods receipt, shipping and RMA logistics | |||||
| Asset register ownership and audit | |||||
| Photographic evidence on every change | |||||
| Round-the-clock infrastructure monitoring | |||||
| Incident ownership through to closure | |||||
| OEM and warranty escalation management | |||||
| Named service delivery manager | |||||
| Live DCIM for power, thermal and capacity | |||||
| On-site spares pool held and managed | |||||
| Committed availability target |
Included as standardAvailable as an optionAny capability can be added to any tier, and the mix can differ site by site.
07 / The systemsWhat you actually log into
Every tier includes the portal: raise and track tickets, interrogate the asset register and pull service reports on demand. From the upper tiers we instrument the estate itself, so power, thermal and capacity arrive live in DCIM rather than being asked for. These are our own systems, built in house rather than licensed in, which is why they can be shaped around the way you already work.
Live previews of the real systems, running in the page. Select one to see what it is used for.
We are not precious about tooling. Run ours on its own; keep what you already have and we operate it; or run both, with two-way ticket sync over API so an event on the floor opens, updates and closes in your system without anyone retyping it. Taking on an incumbent’s tooling is a normal part of a transition, not an exception to it.
08 / The allowanceHow much downtime your target allows
Availability targets are quoted far more often than they are costed. What the figure buys is a fixed allowance of unplanned downtime a year, and every minute of it is spent whether someone is travelling to site or already working on the fault. It is worth seeing the numbers written down before agreeing to one.
Pick a target, then check each tier against it
Unplanned downtime allowed per year. A tier fits when its response clock is inside the allowance; diagnosis and repair still have to come out of what is left.
Response clock against the allowance
Allowance 8 hr 46 min a year
Planned maintenance is excluded: most availability targets are written against unplanned outage only. Check which yours means before committing to it.
A response clock is spent before any diagnosis starts. Around three nines a same-day attendance still fits inside a single outage. By four it does not: the travel alone would spend the year’s budget, so the only way to buy the response is to have somebody already on the floor. Beyond that, availability stops being an operations question at all. Five nines is about five minutes a year, won with redundancy and concurrent maintainability; operations is what keeps it there.
09 / Getting startedThe two ways an operations contract begins
Whoever you appoint, there are only two starting positions: an estate that does not exist yet, and one that is already running. They are different pieces of work, and the second is the one most often underestimated.
Deploy and commission
Site selection, goods receipt, structured cabling, racking or a migration of existing kit, labelling, staged power-on, burn-in, validation and a formal handover into steady state. Normally priced as a project rather than a subscription. The advantage is that whoever runs it afterwards already knows every cable in it.
Survey, integrate, take over
A physical survey against the existing records, a baselined asset register, monitoring and DCIM connected, then the runbook, escalation path and SLA formally assumed. Nothing is rebuilt and nothing stops. The work is in the record: almost every live estate has drifted from its documentation, and the audit is what makes the SLA safe to sign.
If you are moving between providers, the transition audit is the part to scrutinise. An incoming operator who accepts an inherited asset register without verifying it is accepting your drift as their baseline, and the first incident is where that gets discovered.
10 / How Optronix helpsOne model, either end of the ladder
Operations is the longest stage of the data centre lifecycle, the years between commissioning a facility and decommissioning it. We run the model set out above at either end: called to site only when we are needed, or a full Optronix crew on your floor around the clock owning the whole operation. Five tiers on presence and depth, one portal, one ticket queue and a named service manager, whether we built the floor or inherited it.
Your presence on site
Not remote hands sold by the fifteen minutes. We become the hands on the rack, the eyes on the estate and the system of record for every asset in it, so your team never needs to enter the facility.
Any tier, any number of sites
From attendance on request to a dedicated crew on the floor around the clock. Scheduled operations are running today across five cities in two continents, and every tier is available anywhere we can reach and staff a site.
The floor and the fabric
Rack-level operations and network operations run on the same ladder and the same queue, so an alert on a switch and a body in the hall are one ticket rather than two suppliers.
Evidence, not assurances
Photographic evidence on every change, a maintained asset register, live DCIM from the upper tiers, and reporting you can hand to your own auditors.
If you would rather see the whole model first, the deck runs through every tier, the capability matrix and the portal, with the systems above running live inside it.
Related services
Managed operations
Tell us the site and the hours that matter.
Tell us the coverage you need and we will come back with the tier that fits, what it commits to and what it costs.
Sources and notes
- Uptime Institute Annual Outage Analysis: human error as a factor in most significant outages, and the planned and unplanned distinction.
- Availability and downtime figures are the standard arithmetic of the nines (99.999% is about 5.26 minutes a year).
- Common industry practice for change control, MOP/SOP/EOP, DCIM, and capacity and asset management.
- The five-tier structure in section 06 is the Optronix service model. Tier names are descriptive; there is no industry standard for them.
Sections 01 to 05 and 08 are general guidance and apply whoever you buy from. They are not a substitute for site-specific engineering or your facility’s own procedures. Downtime figures are the arithmetic of the nines, not a guarantee: real availability depends on design, kit and, above all, operations.