The server has stopped responding, and there’s no one on site at the rack. If you want to restart a server remotely, you need a method of access that works independently of the frozen operating system and doesn’t require a technician on site: the Baseboard Management Controller, or BMC for short.
The BMC runs on the power supply unit's standby power and remains accessible even when the system is turned off or frozen. It can be used to remotely reboot a server, even if the operating system is no longer responding. Redfish and IPMI provide commands for this purpose, such as Graceful Shutdown, Hard Reset, and PowerCycle, in exactly that order.
The BMC remains accessible even if the server stops responding
Every server with a dedicated management processor has what is known as a baseboard management controller (BMC). The BMC is located on the motherboard, has its own network interface, and operates independently of the main processor. On the market, this component is known by different names depending on the manufacturer: Dell calls it iDRAC, HPE calls it iLO, Supermicro sticks with the abbreviation BMC, and Raritan offers external KVM-over-IP devices as an alternative for servers without manufacturer-specific BMC access. The BMC is connected to the standby rail of the power supply and therefore receives power even when the server is turned off or the operating system has frozen. A kernel panic, a blocked process, or a completely hung system does not affect this. The BMC continues to run as long as the power supply itself is receiving power.
This feature is what makes remote access at the server level possible in the first place. Without BMC, the only option would be to unplug the power cord—with all the risks that entails for file systems and running processes. With BMC, a server can be managed remotely using two parallel protocols: IPMI and Redfish. These commands reach the BMC via the management network, which is ideally separated from production operations. The article „Separating the Management Network: Redundant Architecture for Emergencies” describes how this can be implemented. The principle behind it is explained in more detail in the article Out-of-Band Management.
Restarting a Server Remotely: The Escalation Chain
Redfish is the REST standard used by the DMTF maintains and which now covers most of the newer server generations. It supports clearly named power commands: ForceOff for an immediate hard shutdown, GracefulShutdown for a controlled shutdown via the operating system, GracefulRestart for a restart with running processes being terminated, and PowerCycle for a complete power cycle (turning the system off and then back on). IPMI, the older and still widely used protocol, offers the same basic functions via the command-line tool `ipmitool`, though with a less uniform structure and without Redfish’s more modern authentication options.
During operation, these commands follow a fixed sequence. First comes the graceful shutdown, during which the operating system receives the signal to terminate processes and completely unmount file systems. If the server does not respond within a reasonable amount of time, a hard reset follows—an immediate reboot that ignores any open processes. If that also has no effect, a power cycle is performed: power is completely turned off, followed by a brief pause, then power is turned back on. This is followed by two additional steps, which are rarely necessary but exist in case the server itself is no longer the actual problem.
| Stage | Action | Effect |
|---|---|---|
| 1 | Graceful Shutdown | The operating system shuts down in a controlled manner |
| 2 | Hard Reset | Immediate restart regardless of any open processes |
| 3 | PowerCycle | Turn the power off completely and then back on |
| 4 | BMC Reset | The management processor restarts on its own |
| 5 | PDU | Switchable power strip cuts off the power supply from the outside |
Why Order Matters
A hard reset works quickly—and that's exactly what makes it so tempting when a server stops responding. Still, there's a reason for the difference between the two levels; it's not just a formality.
If you go through the steps from top to bottom instead of immediately resorting to a hard reset, the worst that can happen is that you’ll lose a few minutes of waiting time. If you skip them, you risk damaging the file system and having to perform a recovery that takes significantly longer than the few minutes it would have taken to follow the proper procedure.
When even the BMC stops responding
It is rare, but not impossible, for a BMC to hang—for example, after a failed firmware update or an internal software error. In this case, a targeted BMC reset via the serial interface will usually resolve the issue. If the BMC also stops responding, the final option is a switchable Power Distribution Unit (PDU). The PDU is located between the power outlet and the server and can be controlled via the network independently of the server and its BMC. It is therefore at the end of the chain. It completely disconnects the server from the power supply without requiring any component within the server itself to be involved.
A look at the alternative shows just how costly it can be to do without these steps. If, instead, someone actually has to travel to the data center, the cost of an on-site visit, according to vsight.io ranging from $150 to $1,000, depending on travel time, wait time, and technician availability (vsight.io, 2026). This range comes from a U.S. source and cannot be applied directly to German or Austrian conditions. Nevertheless, it serves as a rough estimate of the effort saved by a functioning remote access solution.
Estimated cost of a single on-site service call, U.S. source.
vsight.io, 2026
Establish Your Own Escalation Process in Advance
A frozen server rarely follows the pattern of the last failure. If you define the five steps for your own fleet in advance, you won't have to stop and think about what command to issue next when an emergency arises.
Frequently Asked Questions
What should you do if the BMC freezes too?
If the BMC itself no longer responds to Redfish or IPMI commands, a switchable PDU is usually the only solution. It cuts off the external power supply, regardless of whether the BMC is still functioning. Serial access via the console may allow you to perform a targeted BMC reset beforehand without having to reboot the server itself.
Is a hard reset dangerous?
A hard reset is not inherently dangerous, but it can corrupt data if write operations are in progress, because file systems and databases are not given the opportunity to shut down properly. That is why a graceful shutdown—which gives the operating system this opportunity—is always the first step in the escalation chain.
Do you need a switchable PDU for this?
A switchable PDU isn't strictly necessary; a BMC already handles most failures via Redfish or IPMI. It makes sense as a last line of defense in the rare event that the BMC itself stops responding and an on-site visit would otherwise be unavoidable.
torck develops its own software that allows a server to be restarted remotely—from a graceful shutdown to a switchable PDU—and initially runs it in its own test environment. The teams in Maxhütte-Haidhof, Vienna, and Rabat manage their own servers using this system and therefore know from firsthand experience where a reset command might fail. The contractual partner is the German company torck GmbH; this is a contract without a chain of subcontractors. If you want to see how KVM Fleet If this escalation chain applies to your own server fleet, schedule a product demo or an initial consultation—with no obligation and no sales pressure.
German 
English 
French