Platform
The parts you only think about when they fail.
Hardware, network, storage and the operational habits behind them. This page is the detail that does not fit on a pricing card, written for the person who will be on the other end of a pager.
Compute
KVM on Xeon, sized honestly.
Servers run on Intel Xeon and AMD EPYC hypervisors under KVM. Each guest gets its own kernel and its own virtual block device. Nothing is a container dressed up as a virtual machine.
- Allocation ratios are bounded. We cap how many vCPUs a host can sell against its physical cores, and we watch steal time per host rather than waiting for tickets.
- Memory is not overcommitted. Ballooning is off. The gigabytes on the invoice are the gigabytes in the guest.
- Nested virtualisation is on. Docker, Podman and k3s work without workarounds.
- Live console access. VNC reaches the guest independently of its networking, so a bad firewall rule is not a rebuild.
Network
Blended transit plus local peering at every site. The point of running in 40 countries is short paths, and short paths come from peering locally rather than hauling traffic to a hub.
Local peering first
Each site connects to the regional exchange — DE-CIX in Frankfurt, AMS-IX in Amsterdam, HKIX in Hong Kong, IX.br in São Paulo, UAE-IX in Dubai. Domestic traffic stays domestic instead of touring a continent.
Multiple transit providers
Every facility takes transit from at least two of the global carriers so that one provider having a bad day changes your latency rather than your availability.
Edge filtering
Volumetric attacks are absorbed upstream where the capacity exists. Thresholds vary by country with the carrier mix; ask us for the specifics of a site before you commit.
NVMe storage
Enterprise NVMe in a redundant array per hypervisor, with power-loss protection. The capacity on the plan is the usable size of your device.
Weekly off-host images
A full disk image every week, written to separate storage. Restore from the panel. Daily retention is available as an extra where a week is too long a window.
Access control
Panel accounts support two-factor authentication and per-server API tokens. Console sessions are logged. Support will not change credentials on an unverified request.
Operations
What happens when something goes wrong.
Every provider has incidents. What varies is whether you find out from the provider or from your own users.
- Planned work: 72 hours of notice. By email to every affected customer, scheduled in the local overnight window.
- Incidents get told, not hidden. We write to affected customers while we are still working on it, not after we have a tidy story.
- Failed hardware is migrated, not debated. A dying hypervisor means your server moves; we tell you afterwards rather than asking first.
- SLA credits are claimable. The terms are written down and the credit is a number, not a goodwill gesture.
Support hours
Tickets are answered around the clock, every day, from Dubai. Business-hours cover spans the Gulf, Europe and most of Asia in a single working day; the overnight rota covers the Americas.
First response targets are one hour for a service-affecting incident and four hours for everything else. They are targets we publish because we intend to be measured on them.
Reaching a human
Tickets through the panel are fastest because they arrive with your server attached. Email works too: support@flowvds.com.
Platform questions
The port is not shaped, so a single server can saturate it in a burst. Sustained line rate for weeks is a different conversation: the uplink is shared with other tenants on the hypervisor, and we will contact you before anyone gets throttled. In practice, sustained transfer in the hundreds of megabits is normal and nobody notices.
Volumetric attacks are filtered upstream at the facility edge, which is where the capacity to absorb them exists. Coverage and thresholds differ by country because they depend on the carrier mix at each site. Application-layer attacks are yours to handle on the server — we will help you read the logs and tune the rules.
Enterprise NVMe in a redundant array on each hypervisor, with power-loss protection. The quoted capacity is the usable size of your block device, not a thin-provisioned promise. IOPS are not sold as a separate product and are not artificially capped below what the array can serve.
Yes. The weekly image is written to separate storage in the same facility, so losing a hypervisor does not lose the backup with it. It is a disaster-recovery mechanism, not an archive: it is not a substitute for your own off-site backups of anything you cannot lose.
Planned work that requires a reboot is announced at least 72 hours ahead by email to every affected customer, and scheduled in the local overnight window. Emergency work — a failing drive, an urgent hypervisor security patch — happens when it has to, with notice as soon as we have it.
Test it against your own network first.
Ask for a test IP and a test file in any of the 40 countries. Measure it yourself before you spend anything.