Load profile
A week of readings from the existing servers: processor, memory, disk operations, genuine peaks. Cores, memory and storage are then calculated with three years of headroom.
Virtualisation is sold as a saving on hardware. What you are really buying is time: the time to undo an update that went badly, the time to bring a server back up elsewhere while the first one is dead, the time to build a test machine in ten minutes instead of a week. Five boxes of different ages, each running at fifteen per cent, become one or two hosts on which every server is a file. Since the licensing changes at VMware, many French companies are also looking at Proxmox and Hyper-V. We compare them without loyalty to a brand, and the whole job is done through remote consoles.
Sizing comes from measurements, never from a vendor catalogue. That is what stops a host being too small within a year or twice the size you needed.
A week of readings from the existing servers: processor, memory, disk operations, genuine peaks. Cores, memory and storage are then calculated with three years of headroom.
Local NVMe, ZFS, Ceph or a shared array: the choice follows from the number of hosts and, above all, from how much time you are willing to lose after a failure.
Separate networks for management, guests and replication, so that copy traffic never smothers the people trying to work.
Machines cross one at a time, each in a short slot outside office hours, with the old box ready to be switched back on if anything looks wrong.
Prepared Windows Server and Linux images, patched and already monitored. A test server comes from a template rather than from an installer.
Replication to a second host or to a hosting provider, plus a written procedure: what starts first, on which addresses, and who tells the users.
The timetable depends on how many servers there are and on the lead time of hardware you order through your own reseller.
A week of readings and a conversation about what is genuinely critical and what could simply be retired.
A precise configuration for your supplier to quote, who installs it in your office or in a data centre.
Once the machine has power and a network, we come in through the management console, lay down the hypervisor and networking, then measure what it delivers.
Servers move one by one, the checks are done with the people who use them, and the old hardware is powered down only afterwards.
A snapshot is not a backup. The one taken just in case before an update should live for a few hours, not six months: after that it grows, drags the machine down and eventually fills the datastore, usually on a Friday. So forgotten snapshots are among the first things we list when taking over an environment, and we put an alert on their age.
Sometimes, for a reason that has nothing to do with saving on hardware: being able to roll back, and to rebuild a machine on any compatible hardware. If both servers are recent, properly backed up and a day of downtime costs you very little, wait for the renewal cycle. We will say so when that is the case.
It is the norm now, French publishers included. The condition fits in one sentence: the database needs fast disks and reserved memory instead of fighting the file server and the backup job for them. Before migrating we time a handful of everyday operations, closing an accounting period for instance, and compare afterwards.
The answer is a three-year calculation including licences and support, and it turns on two details people forget: whether your backup software can talk to the new platform, and who on your side will administer it. We move one machine that carries no risk first, let it live for a few weeks, and only then does the rest follow, one per maintenance window.
Yes, and for many companies it is the better compromise: redundant power, cooling and connectivity are included, and your data stays in France or elsewhere in Europe with OVHcloud, Scaleway or a comparable provider. Weigh it against the true cost of a room on your premises, UPS and power cuts included. Your office internet link then becomes the part to make resilient.
Not always as they stand. An agent installed inside the operating system keeps doing its job, but you lose the main prize: a copy of the whole machine, restorable in one move. So we use the project to revisit the backup plan, with an off-site copy and a rehearsed restore before we call the migration finished.
List the servers you run today and what each one does. We reply with a starting configuration and a rough figure for licensing.
Message received
An answer follows inside one working day. Report an outage that is stopping people working and it moves ahead of everything else.
Nothing here under that name. Check the spelling, or simply choose the nearest large city instead. Since every engagement runs remotely, whichever you pick changes nothing about what we do for you.