20.04.2022

Cause of burnout: Investigation of OVH data center fire

The investigation of the fire that destroyed OVH's Strasbourg Data Center (SBG2) in March 2022 concluded that the unfortunate event was a result of catastrophic violations of fire safety requirements.

Konstantin Nagorny, Chief Engineer at Linxdatacenter, St. Petersburg 

The investigation of the fire that destroyed OVH's Strasbourg Data Center (SBG2) in March 2022 concluded that the unfortunate event was a result of catastrophic violations of fire safety requirements.  

Here we take a closer look at the circumstances of the fire and make a list of dos and don'ts for the future.

The fire at VH's Strasbourg Data Center (SBG2) led to the blackouts of 3.6 million sites and services across the global web. The shutdown affected government portals, banks, retailers, and media.

Despite the efforts of firefighters, the whole building was lost in a large uncontrollable fire. The circumstances of the fire remind us that even the major providers’ infrastructure is prone to design mistakes and can be vulnerable to disasters.

The first versions and the final report 

The early findings from a preliminary investigation pointed to the uninterruptible power supply (UPS) devices as the cause of the fire. According to media reports, they caught fire a few hours after maintenance that included replacement of a few parts.

The UPS devices reportedly caught fire again several days after the main fire was gone. It’s highly likely that the data center owner used lithium batteries, which are almost impossible to extinguish due to the exothermic chemical reactions caused by the extreme heat. Reignition in the off state confirms this hypothesis.

The fire brigade of the Bas-Rhin department that worked on the site reported electric arcs longer than 1 m in the UPS room.

Due to the lack of a main switch, it took around 3 hours to turn off the power supply of the data center. The UPS inverters were energized for all that time and some servers continued to work.

In addition, the electrical cable routers were not isolated and the floors and ceilings of the data center were made of wood. The free cooling system which normally supplied the servers with cool fresh air helped the fire to spread across the building.

Finally, SBG2 had no automatic fire extinguishing system that could have mitigated the disaster.

Let’s take a closer look at how all these details led to the massive fire, and how it could possibly have been prevented.

Wooden floors and ceilings 

A data center should be designed as a complex of isolated compartments with walls, doors, and service shafts constructed of fire-resistant materials. They should be able to resist fire for 45 to 60 minutes.

All these measures should provide enough time to evacuate personnel, call firefighters, and power off the equipment. Yet this is no panacea: the temperature may be high enough to burn the metal. That’s why some countries impose even higher standards and require all metal constructions to be additionally covered with a refractory paint.

Most likely, the wooden components of the data center had some coating, but a low-quality one. Additionally, the SBG2 spaces weren’t really isolated compartments, which allowed the fire to spread quickly from area to area and eventually destroy the entire building.

Lack of automatic fire extinguishing system 

There’s currently no generally accepted standard for automated extinguishing in data centers. Most often, everything is up to the owner.

Yet it’s mandatory for a data center to have an automated fire alarm system that can detect a fire and launch processes that prevent it from spreading and causing further damage.

Even in the absence of an automated fire extinguishing system, there are solutions that compensate for it. For instance, the owner can reduce the use of flammable materials in some critical areas in the data center.

Spread of fire due to free cooling  

After the fire alarm goes off, the ventilation system should be shut down, and fire dampers between compartments should be closed. The dampers are located at the edge of each compartment and prevent fire from spreading, as well as cut off oxygen supply from the outside.

In case of fire, a ventilation system functioning as normal can be a serious factor that helps the fire to spread.

Emergency power-off (EPO) 

A kill switch, or emergency power-off, is a safety mechanism to power off the whole facility in case of emergency when the power can’t be turned off in the usual manner.

Some people think that a single EPO switch can be dangerous in terms of operational sustainability, since if it is pressed by accident, it can power off the whole data center. Dedicated kill switches for different data center systems can be a more reliable solution. For instance, separate switches can be used for separate lines of UPS that power the servers, reducing the risk of simultaneous shutdown of two UPS beams in case of false alarm or accidental trigger.

The EPO buttons must be located outside the equipment rooms. Employees need safe access to them in case of fire. If no special EPO boards are installed, an ordinary electrical switch can work, if located on a switchboard outside the equipment room.

Conclusions 

There are several measures to consider to improve the fire safety of a data center.

The premises of the data center should be built as isolated fire compartments. The walls, the floors, and the ceiling should be fireproof to prevent fire from spreading in any direction. Even if the insides of a data center are designed as isolated fire compartments, a roof made of flammable materials will make it extremely vulnerable to fire. In case of fire, the flames will spread all over it, and firefighters will have to flood the entire data center with water to extinguish it.

The cable routes through the walls of fire compartments should be filled with non-flammable materials with the same level of fire resistance as the walls. For instance, a special foam can be used. Both power cables and telecom cables should have fire isolation to prevent fire spreading over them like a fuse.

Cable routes and electrical switches should be in good working condition. Electrical switches and EPO buttons should be located outside the equipment rooms and be safe to use to power off the burning equipment.

The data center should have an automated fire alarm system that takes the design of the server rooms into account. For instance, to detect smoke in a large volume of fast-moving air, special early smoke detection sensors should be used.

For premises where diesel generators are located, IR sensors should be used to detect fire in the IR spectrum to prevent false alarms from the regular sensors that detect heat and smoke. In the noisy environment of server rooms with UPSs and DGUs, stroboscopic alerters should be used to warn the employees in time.

If a fire alarm is triggered, the ventilation should be automatically turned off and the fire dampers should be closed to isolate fire compartments and stop the supply of oxygen to the premises. A data center with a closed cooling system can continue operations in case of a local fire. A free cooling system will be shut completely once the ventilation is locked.

It’s a good idea to install gas fire extinguishing systems in some critical areas of a data center, such as server rooms, UPS rooms, and battery rooms. The gas should be safe for humans to prevent health damage for the employees if they find themselves trapped in the premises by fire.

Fires in data centers are usually caused by faulty electrical wiring, electrical equipment and switches, hot welding and other uses of open flames. It’s important to minimize the risks of man-made fire by documenting and monitoring all the work. Of course, smoking should be prohibited inside the data center.

Storage of any flammable materials in the data center should be organized outside the server rooms and other critical areas.

We hope this article helped you to learn more about fire safety in data centers.

 

News and publications

Article
09.08.2022
IS in scarcity: a big transition strategy
News
01.08.2022
Linxdatacenter is in the TOP10 of the largest DC service providers
News
25.07.2022
Linxdatacenter launches its own PaaS tools
News
20.07.2022
St. Petersburg cloud Linxdatacenter passed the certification on ...
Article
30.06.2022
How we optimized customer data center management
News
27.06.2022
Linxdatacenter: Russian Cloud Market to Grow in 2022...
News
26.05.2022
Anna Malmi has been appointed the regional director of Linxdatacenter...
Article
20.05.2022
Cloud Edge: What's Happening in the Russian Market - Linxdatacenter
News
13.05.2022
The new CEO of Linxdatacenter is Andrei...
Article
03.05.2022
Unit per monoblock: the modular UPS revolution in data centers

You may also be interested in

Linx Outsourcing
Audit, upgrade and optimize your server capacities
read more
Data-center management outsourcing
Linx Network
Ensure network resiliency and uptime
read more
Network services
Linx DraaS
Protect your IT systems today!
read more
Disaster Recovery as a Service

Write to us

How we optimized customer data center management

Data center is a complex IT and engineering object, which requires professionalism at all levels of management: from managers to technical specialists and executors of maintenance works. Here's how we helped our client put operational management in corporate data centers in order.
 

Taras Chirkov, Head of Data Center in St. Petersburg  in St. Petersburg 

Konstantin Nagorny, chief engineer of data center in St. Petersburg.  in St. Petersburg 

Data center is a complex IT and engineering object, which requires professionalism at all levels of management: from managers to technical specialists and executors of maintenance works. Here's how we helped our client put operational management in corporate data centers in order.  

Management is in the lead 

The most advanced and expensive IT equipment will not bring the expected economic benefits if proper processes of engineering systems operation in the data center, where it is located, are not established.  

The role of reliable and productive data centers in today's economy is constantly growing along with the requirements for their uninterrupted operation. However, there is a big systemic problem on this front.  

A high level of "uptime" - failure-free operation of a data center without downtime - depends very much on the engineering team that manages the site. And there is no single formalized school of data center management.  

And there is no single formalized school of data center management.    

Nationwide  

In practice, the situation with the operation of data centers in Russia is as follows.  

Data centers in the commercial segment usually have certificates confirming their management competence. Not all of them do, but the very specifics of the business model, when a provider is responsible to the client for the quality of service, money and reputation in the market, obligates them to own the subject. 

The segment of corporate data centers that serve companies' own needs lags far behind commercial data centers in terms of operational quality. The internal customer is not treated as carefully as the external customer, not every company understands the potential of well-configured management processes. 

Finally, government departmental data centers - in this regard, they are often unknown territory due to their closed nature. An international audit of such facilities is understandably impossible. Russian state standards are just being developed.  

This all translates into a "who knows what" situation. "Diverse" composition of operation teams composed of specialists with different backgrounds, different approaches to the organization of corporate architecture, different views and requirements to IT departments.  

There are many factors that lead to this state of affairs, one of the most important is the lack of systematic documentation of operational processes. There are a couple of introductory articles by Uptime Institute which give an idea of the problem and how to overcome it. But then it's necessary to build the system by your own efforts. And not every business has enough resources and competence for that.  ⠀⠀  

Meanwhile, even a small systematization of management processes according to industry best practices always yields excellent results in terms of improving the resilience of engineering and IT systems.  

Case: through thorns to the relative order 

Let's illustrate by the example of an implemented project. A large international company with its own data center network approached us. The request was for help to optimize the management processes at three sites where IT systems and business-critical applications are deployed.  

The company had recently undergone an audit of its headquarters and received a list of inconsistencies with corporate standards with orders to eliminate them. We were brought in as a consultant for this as a bearer of industry competence: we have been developing our own data center management system and have been educating on the role of quality in operational processes for several years.  

Communication with the client's team began. The specialists wanted to get a well-established system of data center engineering systems operation, documented on the processes of monitoring, maintenance and troubleshooting. All this had to ensure optimization of the infrastructure component in terms of IT equipment continuity.  

And here began the most interesting part.  

Know thyself 

To assess the level of data centers in terms of compliance with standards, you need to know the exact requirements of the business to IT systems: what is the level of internal SLA, the allowable period of equipment downtime, etc.  

It became clear right away that the IT department did not know exactly what the business wanted. There were no internal criteria of service quality, no understanding of the logic of their own infrastructure.  

Colleagues simply had no idea what the permissible downtime for IT-related operations was, what the optimal system recovery time in case of a disaster was, or how the architecture of their own applications was structured. For example, we had to figure out whether a "crash" of one of the data centers would be critical to the application, or if there were no components affecting the application.  

Without knowing such things, it is impossible to calculate any specific operational requirements. The client recognized the problem and increased coordination between IT and the business to develop internal requirements and establish relationships to align operations.  

Once an understanding of the IT systems architecture was achieved, the team was able to summarize the requirements for operations, contractors, and equipment reliability levels.  

Improvements in the process 

Our specialists traveled to sites to assess infrastructure, read existing documentation, and checked the level of compliance of data center projects with actual implementation.  

Interviews with the responsible employees and their managers became a separate area of focus. They told what and how they do in different work situations, how the key processes of engineering systems' operation are arranged.  

After starting the work and getting acquainted with the specifics of the task the client "gave up" a little: we heard the request "just to write all the necessary documentation", quickly and without deep diving into the processes.  

However, proper optimization of data center "engineering" management implies the task to teach people to properly assess the processes and write unique documentation for them based on the specifics of the object.  

It is impossible to come up with a working document for a specific maintenance area manager - unless you work with him at the site continuously for several months. Therefore this approach was rejected: We found local leaders who were willing to learn themselves and lead their subordinates.  

Having explained the algorithm of documents creation, requirements to their contents and principles of instructions ecosystem organization, for the next six months we controlled the process of detailed writing of documentation and step-by-step transition of the personnel to work in a new way. 

This was followed by a phase of initial support for work on the updated regulations, which lasted one year in a remote format. Then we moved on to training and drills - the only way to put the new material into practice.  

What's been done 

In the process, we were able to resolve several serious issues.  

First of all, we avoided double documentation, which the client's employees feared. To this end, we combined in the new regulations the regulatory requirements applied to various engineering systems as standard (electrics, cooling, access control), with industry best practices, creating a transparent documentation structure with simple and logical navigation.   

The principle of "easy to find, easy to understand, easy to remember" was complemented by the fact that the new information was linked to the old experience and knowledge of the employees. 

Next, we reshuffled the staff of service engineers: several people turned out to be completely unprepared for the change. The resistance of some was successfully overcome in the course of the project through the demonstration of benefits, but a certain percentage of employees turned out to be untrained and unresponsive to new things.  

But we were surprised by the company's frivolous attitude to its IT infrastructure: from the lack of redundancy of critical systems to the chaos in the structure and management.  

In 1.5 years the engineering systems management processes have been pumped up to the level that allowed the company's specialists to successfully report "for quality" to the auditors from the headquarters.  

With the support of the operating component development pace the company will be able to pass any existing certification of data centers from leading international agencies.  

Summary 

In general, the prospects of consulting in the field of operational management of data centers, in our opinion, are the brightest.  

The process of digitalization of the economy and the public sector is in full swing. Yes, there will be a lot of adjustments in the launch of new projects and plans for the development of old ones, but this will not change the essence - the operation should be improved at least to improve the efficiency of already built sites.  

The main problem here: many managers do not understand what thin ice they are walking on, not paying proper attention to this point. The human factor is still the main source of the most unpleasant accidents and failures. And it needs to be explained.  

Government data center projects are also becoming more relevant now and require increased attention in terms of operations: the scope of government IT systems is growing. Here, too, the development and introduction of a system of standardization and certification of sites will be required.  

When the requirements to public data centers in Russia at the level of legislation will be reduced to a standard, it can be applied to commercial data centers, including for the placement of public IT resources.  

The work in this area is ongoing, we are participating in this process in consultation with the Ministry of Digital and by building competencies for teaching courses on data center operation at the ANO Data Center. There is not much experience on such tasks in Russia, and we believe that we should share it with colleagues and clients. 

Cause of burnout: Investigation of OVH data center fire

BEST, money transfer and payments operator

business challenge

The customer faced a technical issue with a persistent BGP session flag with Linxdatacenter hardware. We examined the problem and found out that one of customer’s hosts was under a DDoS attack.

Because of the distributed nature of the attack, traffic couldn’t be filtered effectively, and disconnecting the host from the external network wasn’t an option. The attack stopped after changes in the server configuration, but resumed the day after. A 5.5 Gbps attack overloaded the junctions with internet providers, affecting other Linx Cloud users. To mitigate the effects of the attack, we employed a dedicated DDoS protection service.

Solution

To ensure the continuous availability of resources hosted in Linx Cloud, we rerouted all the customer’s traffic through StormWall Anti-DDoS system. The attack was stopped within half an hour. To prevent future cyberattacks, we organized all connections to the customer’s resources through the StormWall network.

client:

BEST, money transfer and payments operator

business challenge

The customer faced a technical issue with a persistent BGP session flag with Linxdatacenter hardware. We examined the problem and found out that one of customer’s hosts was under a DDoS attack.

Because of the distributed nature of the attack, traffic couldn’t be filtered effectively, and disconnecting the host from the external network wasn’t an option. The attack stopped after changes in the server configuration, but resumed the day after. A 5.5 Gbps attack overloaded the junctions with internet providers, affecting other Linx Cloud users. To mitigate the effects of the attack, we employed a dedicated DDoS protection service.

Solution

To ensure the continuous availability of resources hosted in Linx Cloud, we rerouted all the customer’s traffic through StormWall Anti-DDoS system. The attack was stopped within half an hour. To prevent future cyberattacks, we organized all connections to the customer’s resources through the StormWall network.

Thank you for your inquiry, we will get back to you shortly!