Flaem

Dedicated server management: from machine inventory to client console

You have physical servers in your racks and want to rent them out without a call to the data centre every time someone needs a reboot. Flaem takes every machine through its whole life: intake into inventory, a burn-in under full load, instant sale, a network install of the operating system, control through the management controller and a safe return to the shelf when the contract ends.

Flaem: store, billing and client panel
One rack in one locationPDU
Flaem footholdcontrollers and network installs
U02in use
U03in use
U04available
U05service
U06failed
U07in use
One rack in one location. Every location has its own Flaem foothold.
01Machine inventory

Machine inventory: every server with its location, rack and state

Before you sell anything, you add the machine to inventory. From then on the panel knows where it stands, what it is made of and whether it may go to a client.

Machine record in inventory

Location and rack
Data centre, rack and position in the rack. Clients pick the location when they buy.
Serial number
The number from the chassis, so a technician at the rack knows which machine is in front of them.
Network cards
Recorded by MAC address. The network install recognises the machine by them.
Controller
Redfish or IPMI. The login details stay in the Flaem control plane.
Hardware profile
CPU, memory, disks and network port. The profile ties the machine to a package in the store.
Network boot
Handled by Flaem, or set by hand when you boot the machine your own way.

What the client sees in the store

  • CPU model
  • Cores and threads
  • Memory
  • Disks
  • Network port
Free machines per locationfreesold
Location A
Location B
Location C

When a location has no free machine, the package cannot be bought there. Nobody pays for a server that is not in the rack.

Five machine states

  • AvailableChecked, with addresses, ready to sell.
  • InstallingThe OS installs over the network for a new client.
  • In useRunning for a client, with renewals and invoices.
  • ServiceOff sale: new, returned or under repair.
  • FailedSomething broke and the machine waits for a technician.

A new machine starts in service. It goes on sale only after it passes the burn-in, gets its addresses and you mark it as available.

02Burn-in

Burn-in before sale: a few hours of every new machine under full load

A faulty memory module or a disk about to die shows up during the test, not in the client's first week. Flaem loads every new machine and keeps a report with the numbers.

Test startsA few hours later
Memory
memtester

Writes and reads patterns across memory to catch damaged modules.

CPU
stress-ng

Loads every core and reads the temperatures, so you see whether cooling keeps up.

Disks
fio and SMART

Writes and reads across the whole surface of each disk, and SMART readings before and after show whether anything got worse.

Network
iperf3

Measures throughput to another node in your network.

Verdict after the testPassed

Burn-in report

Each test records its numbers and the report ends with a single verdict. After a pass you assign addresses and mark the machine as available.

Passed: the machine can go on the shelf
Failed: the machine stays in service
03Instant sale

Instant sale with automatic machine assignment

The client buys a package, not a specific box. Flaem picks a free machine with the package hardware in the location the client chose and starts the install straight away.

1Package and locationThe client picks a package, a location and an OS, and adds their SSH key.
2Payment and reservationA free machine is reserved in the same transaction as the payment.
3Network installThe OS installs in the background, without your team.
4Server deliveredThe client logs in with their SSH key and runs the machine from the panel.

Two orders for the last machine

Reservation and payment are one operation, so the same machine cannot be sold twice. The second buyer is refused before paying anything.

First order
Second order
Last free machinewaiting for an order

An order is stopped before any money is taken when

  • there is no free machine in the chosen location
  • the package has no hardware profile or OS image
  • the client has not added an SSH key

Your price

You set the price of every machine in inventory. Discounts happen only by hand, the panel never marks anything down for you.

With or without commitment

With commitment switched on, a dedicated server contract runs for at least 6 months and the client cannot leave after the first one. You can price longer terms lower.

Ready machines

You sell what stands in the rack and passed the burn-in, not hardware assembled to order.

04Network install

Network OS install and a Flaem foothold in every location

Nobody has to walk to the rack with a USB stick. The machine boots from the network, fetches the installer and gets a fresh OS, while a Flaem foothold in the same location does all the talking to the hardware.

Flaem control planepanel, store and controller logins
internet
Management network in one location
Flaem footholda node on the management network
Machine controllersRedfish or IPMI
Machinesboot from the networkDHCP->TFTP->iPXE

No route into the controller network

The control plane talks to controllers through the foothold, so it needs no VPN and no route into the private management network.

Logins stay in the control plane

The foothold receives controller credentials for a single call and does not store them.

DHCP, TFTP and iPXE on site

The same foothold serves the network install, so you do not run a separate DHCP server for dedicated machines.

One per location

Machines in different data centres or buildings sit on separate management networks, so each location gets its own foothold.

Operating systems to install

  • Ubuntu
  • Debian
  • Rocky Linux
  • AlmaLinux
  • Red Hat Enterprise Linux
  • Fedora

You attach images to the package and the client picks from that list at purchase and at reinstall. A reinstall asks for confirmation because it erases the disks.

05Controller

Control through the BMC: Redfish for newer machines, IPMI for older ones

The management controller is a small computer built into the server that keeps running even when the OS is down. Flaem talks to it on the client's behalf, so a reboot or a power cut is a button in the panel, not a support ticket.

iDRACDell
iLOHPE
XClarityLenovo
SupermicroSupermicro controllers
OpenBMCopen controller firmware
IPMIOlder machines without Redfish are handled over IPMI with ipmitool.
Over Redfish / IPMI

The client, in their panel

  • Power on and off
  • Reboot
  • Hard reset
  • Power cut
  • OS reinstall
  • Rescue system
  • Reverse DNS for addresses

Flaem, through the controller

  • Machine health
  • Sensor readings
  • Controller event log
  • One-time network boot

A PDU for when the controller goes quiet

You assign the machine to an outlet on a remote power strip. If the controller itself hangs, you cut and restore power to that outlet from the panel, without a trip to the data centre.

A hung node gets a hard reset through its neighbour in the same rack.

outlet:power the outlet off and on
06Console

SOL and KVM console in the browser, without handing over the controller

When the OS will not boot or the client has locked themselves out of the network, they still see their machine in the browser. They get a text console from the serial port and the screen image, as if they stood at the rack with a monitor and keyboard.

Client browser
Flaemcontroller logins stay here
Foothold
Controller
Client machine
text from the serial portscreen image and keyboard

Text console (SOL)

The machine serial port passed through the controller. Enough for Linux without a desktop, the boot menu and kernel messages.

Screen image (KVM)

What a monitor plugged into the server would show, keyboard included. Handy in BIOS settings and installers.

Rescue system

The machine boots from the network into a rescue system and the client disks stay untouched. The client fixes the configuration, copies files off and goes back to a normal boot.

The controller stays yours

The console runs through Flaem. The client never gets the controller login or address, because access to the controller would give them power over the whole machine, not only their OS.

07Telemetry

Dedicated server telemetry: controller sensors, system resources and network traffic

The client sees more than "up or down". They get a history of temperatures, fans and power draw from the controller, and with the agent on the machine also CPU, memory, disks and network.

reinstallreboot
Temperaturesfrom the controller, every minute
Fansfrom the controller, every minute
Power drawfrom the controller, every minute
CPUfrom the agent on the machine
Network trafficfrom the port and the agent

Your client sees

Controller sensors
Temperatures, fans and power draw in watts, recorded every minute.
Resources with the agent
CPU, memory and disks when the Flaem agent runs on the machine.
Network
Traffic and packets, with real peaks instead of a smoothed average.

You see

Node hardware health
Disk state, temperatures, memory errors, fans and power supplies, with an alert before something fails.
Percentiles
p50, p90, p95 and p99 for any window, so a short spike does not vanish into the average.
Events on the charts
Reboots, reinstalls, package changes and attacks as markers on the timeline.
08Suspension and return

Suspension, cancellation and a safe return to inventory

The end of a contract is the moment when one client's data could reach the next one. Flaem never lets a machine go back on sale by itself.

Payment missingSuspensionThe controller cuts the power. Client data stays on the disks.
Payment receivedUnsuspensionThe machine powers on by itself if it was running before the suspension.
CancellationService with a notePower is cut and the machine goes to service with a "wipe disks" note.
After the wipeBack on the shelfYou mark the machine as available. Flaem never returns it to sale on its own.

Switch port

For abuse or non-payment you can also shut down the switch port the machine is plugged into, with one click in the panel.

09Operator

Dedicated servers next to games, VPS and networking, on one operator dashboard

Dedicated servers do not need a separate system. You see them on the same dashboard as everything else, bill them with the same invoices and support them with the same team.

Flaem operator dashboard: sales and MRR, live nodes and the issues waiting for a person

One store, one set of invoices

A dedicated server has renewals, reminders, invoices and KSeF e-invoicing like any other service.

Addresses from your pool

IP addresses for dedicated servers come from pools not tied to any node, and the client sets reverse DNS for them.

Team permissions

Who may reboot client machines and who may only browse inventory is set in permissions for every single action.

Licence

Dedicated servers run over IPMI or Redfish are one option: $15 a month for all your dedicated servers, with no server limit. They do not count as nodes, and the number of clients does not matter.

10Questions

Questions about dedicated servers

Controllers, network installs, the console and what happens to a machine when a contract ends.

No. The client has power, reinstall, the rescue system and the SOL and KVM console in the panel, but never gets the controller login or address. The credentials stay in the Flaem control plane.

No. The Flaem foothold in each location serves DHCP, TFTP and iPXE for the dedicated machines there.

They are handled over IPMI with ipmitool. Power, sensors, the event log and network boot work over IPMI just as they do over Redfish.

Ubuntu, Debian, Rocky Linux, AlmaLinux, Red Hat Enterprise Linux and Fedora. You attach images to the package and the client picks from that list at purchase and at reinstall.

Not by itself. After cancellation it goes to service with a "wipe disks" note and returns to sale only when you mark it as available.

Yes. You switch on commitment for the package. The contract then runs for at least 6 months, and you can price longer terms lower.

You cut and restore power to the remote PDU outlet the machine is plugged into. A hung node gets a hard reset through its neighbour in the same rack.

As one option. You pay $15 a month for all your dedicated servers run over IPMI or Redfish, with no server limit. Dedicated servers do not count as nodes, and the number of clients does not change the amount.

We will show dedicated servers on your own machines

Tell us how many servers you have, in how many locations and with which controllers. In the demo we show the inventory, network installs and the console, then work out the licence from the price list.