Availability Sets: Keeping Your Practice Working If a Site Goes Down

No single building has 100% availability, whether it is your office or a hosting site. The best-run sites are built with redundant power, cooling and network, and they still go offline. What matters is whether your staff can keep working if that happens.

Today we are introducing Computle Availability Sets: a second copy of each of your Computle workstations, kept continuously up to date at a second Computle site. If the primary site has an outage, your users move to their replica and carry on working, with their applications, settings and files as they left them.

What an availability set is

An availability set is a pair of machines placed so that no single event can take both out at once.

Your primary workstation runs at your home site as it does today.
A replica runs at a second Computle site, in a different city.
The primary is replicated to the replica continuously, so the replica is always a clone of your working machine.

The replica can be a lower specification than the primary to reduce the cost, as long as it can run your users' work through an outage. We size it with you.

Diagram: a primary workstation replicated over a diverse 100G backbone to a replica at a second site in a different city

Why this is only possible on Computle

Replicating a full workstation continuously needs a lot of private bandwidth between sites. Our UK sites are joined by our own diverse 100G network, running over separate physical routes, and each customer's replication runs over a private backbone dedicated to that customer. Over the public internet, a replica falls behind, and a replica that is hours behind is of little use.

Each workstation has a guaranteed 10 Mbps of replication bandwidth, which bursts up to 1 Gbps when there is more to copy, for example after a large project is saved. This keeps the replica seconds behind the primary in normal use and lets it catch up quickly after large changes.

We also control the entire software stack. Our own management system, Computle VMM, orchestrates machines across all of our sites, which is what lets us keep each replica in step and move users between sites in minutes.

No facility is immune

Outages are common. The Uptime Institute's 2025 survey found that half of all data centres had at least one impactful outage in the past three years, and that power problems caused 45% of incidents. The examples below all involved operators with redundant systems and professional staff.

Date	Operator and location	What happened	Impact
March 2021	OVHcloud, Strasbourg	A fire destroyed the SBG2 building and damaged neighbouring buildings.	An estimated 3.6 million websites went offline. Customer data and backups held in SBG2 were lost.
July 2022	Google and Oracle, London	Cooling systems failed during the UK's 40°C heatwave. Google reported multiple redundant cooling systems failing at once.	An outage of around eight hours.
October 2022	SK C&C, Pangyo, South Korea	A fire in a battery room.	Kakao's services were offline for more than ten hours, including messaging, payments and taxi dispatch for most of the country's 51 million people.
September 2024	Digital Realty, Singapore	A lithium-ion battery fire in the UPS rooms forced an emergency power shutdown.	Alibaba Cloud and several other platforms were disrupted. Firefighters were still on site a day later.
September 2025	National Information Resources Service, Daejeon, South Korea	A UPS battery fire during maintenance.	647 government services were disrupted. 858 TB of data, eight years of work for 125,000 civil servants, was permanently lost because it had no backup.
March 2026	AWS, UAE and Bahrain	Drone strikes damaged three facilities.	More than 100 services were disrupted, affecting banks and payment companies. Six months later, some facilities had still not been restored.

Your office carries the same risk

This is not specific to Computle. A practice with 50 workstations in its own office carries the same risk, and usually more of it. An office can flood, catch fire, be broken into, lose power or its internet connection, or be closed by a police cordon or a terrorist incident nearby. Most have no backup power and no second site.

These outages are not always short. In September 2025, a burst water main in Marylebone, central London, put up to two metres of water into basements on Enford Street and York Street and closed Westminster Magistrates' Court. It took around a year for local businesses to be fully repaired.

Many practices plan to send staff home with laptops and a VPN to a second site. That protects the data, but not the work:

If staff could do their work on laptops, the practice would not have bought desktops.
Large Revit models over home broadband and a VPN are slow to open and sync.
If the office is destroyed, the desktops go with it, and replacing them takes weeks.

With Computle, the laptop is only the screen and keyboard. The workstation is at a Computle site, so losing the office changes nothing. An Availability Set covers the remaining risk: the site the workstations are in.

What an outage costs

Take a 30-person architecture practice with an average charge-out rate of £75 an hour. A lost day is £16,875 in fee-earning time. If a site is lost to a fire, flood or aircraft and it takes three weeks to source and rebuild the machines, that is over £250,000.

One lost working day costs a 30-person practice £16,875; a month of replicas costs £2,844

Why not just have a second set of machines?

You can take a second Computle subscription for every user at another site, but:

An MSP will charge you to manage twice as many machines.
If you turn them off to save money, they are unpatched on the day you need them.
They drift from what your users actually run. Plug-ins and settings added to the primary are missing from the standby.

With live replication, the replica is a clone of the primary. There is nothing extra to patch or manage, and when a user moves to it, it is their machine as they left it.

What happens during an outage

It is fully managed by Computle. There is nothing for your IT team or MSP to set up or run. If the primary site goes down, we move your users to their replicas.

Recovery time (RTO): a few minutes.
Recovery point (RPO): seconds. At most, a user loses the last few seconds of work.

We offer 100% availability across the pair: at any time, at least one of a user's two workstations is available to them.

Pricing

We work to keep our customers' costs down. When US tariffs and DRAM prices raised our costs, we absorbed them and froze our prices (our statement on the price freeze). Asking customers to pay more goes against that, and we have not done it lightly. A 100% availability guarantee needs a second machine at a second site, and that machine has a cost.

You pay for the replica machine, with a 40% discount on its compute and storage because it sits idle unless the primary site is down. We do not charge for replication, and there are no licence or switchover fees. At Computle, a replica is a single flat monthly fee covering the hardware and data centre space, on a three-year term:

Workstation	Standard price per month	As a replica per month
RTX 5050	£123	£73.80
RTX 5070	£158	£94.80
RTX 5080	£209	£125.40
RTX 5090	£301	£180.60

Additional hot storage at the second site is £60 per TB per month instead of £100. For the 30-person practice above on RTX 5070 workstations, replicas cost £2,844 a month, about a sixth of the cost of one lost day.

Doing this yourself with third-party replication software means paying for the software, a second workstation, and space and power at a second site. For an RTX 5070-class workstation:

Cost per workstation per month: Computle £94.80, Azure Site Recovery about £117, Zerto about £165

The second workstation is a £2,300 machine spread over three years. Space and power assumes colocation. A second office costs more.

Getting started

Availability Sets are available to existing Computle customers now, and are optional. Speak to your account manager and we will set out the cost for your practice.

Notes on costs

Computle replica prices are the published three-year-term prices at computle.com/pricing, less 40%.
Third-party software prices are published list prices, converted at about £0.75 to the US dollar. Zerto is sold through partners, and around $1,000 per protected machine per year is the commonly quoted figure. Azure Site Recovery is $16 per protected machine per month when replicating to your own site.
The second workstation is an RTX 5070-class machine (Intel i7, 64 GB RAM, 2 TB NVMe) at about £2,300, spread over three years.
Space and power is an estimate for colocating one workstation: its electricity at typical UK business rates, plus a share of rack space. A second office costs more.
The lost day is 30 staff × 7.5 hours × £75 average charge-out rate. Rates vary by practice.
The third-party figures exclude MSP management and staff time, which add to their cost.