Skip to main content

Build a fault-tolerant infrastructure

This instruction describes how to increase the fault tolerance of your service at the physical level.

Fault tolerance is the ability of a service to remain operational and continue performing business tasks even if individual components of the IT infrastructure fail.

The service's level of fault tolerance depends on the type of tasks being performed. The need to increase fault tolerance arises when the cost of downtime exceeds the cost of ensuring uninterrupted operation. For example, if the service provides continuous access to critical information, or if company operations depend directly on the uninterrupted performance of IT infrastructure.

What Selectel provides

Selectel ensures the fault tolerance of your service at the level of the server room where server racks are located.

Power supply

Power inputs enter the server room, redundant at higher levels (Selectel transformer substations; industrial UPSs including battery arrays; emergency power source — diesel generator sets), then two independent power feeds are supplied to each rack:

  • for servers with a single power supply, ATS (automatic transfer switch) devices are installed in the racks, to which two independent power inputs are connected; if one input is disconnected, power will continue to be supplied through the second;

  • for servers with two power supplies, they are connected to two independent power strip blocks.

Access to private network and the internet

Access to the private and internet networks is reserved at the aggregation switch level and above. In some cases, a single switch is installed for a pair of racks. Read more about how to check the switch in a rack in the instructions Change rack for a dedicated server.

Features of access to the private and internet networks depend on the type of server configuration:

  • for prebuilt configuration servers, access switches for the private network and Internet switches are installed in the racks;

  • for some Chipcore Line servers there is no private network connection, only Internet switches are installed in the racks;

  • for custom configuration servers, you determine the redundancy of connection to the required network (private or Internet) yourself.

Read more about dedicated server networks in the instructions Public networks and subnets and Private networks and subnets.

Rack location

If technically feasible, when ordering two or more servers, they are placed in different racks. You can view the current server location and submit a request to move the server to another rack in the control panel: in the top menu, click Products and select Dedicated servers → Servers → the Server locations tab.

Improve fault tolerance

You can improve the fault tolerance of a client service at the physical level by reducing the number of potential single points of failure. When building an IT infrastructure, potential points of failure can include:

  • the server itself and its components (disks, power supplies, network interfaces, etc.);
  • ATS (automatic transfer switch);
  • internet access switch and switching (copper and optical connection, transceivers, patch cords, etc.);
  • private network access switch and switching.

Improve fault tolerance at the rack level

To reduce the number of potential points of failure at the rack level, you can:

  • choose servers of custom and prebuilt configurations with server processors. In such servers, all components are optimized for high load and smooth operation and are less likely to fail;
  • use a server configuration with two or more power supplies. This power redundancy method is more reliable as it has no single point of failure. Failure of one of the power supplies or the ATS will not cause the server to shut down;
  • for custom configuration servers, reserve connection to access switches of the required network (private or Internet) via MC-LAG;

Improve fault tolerance of a client service consisting of multiple servers

To reduce the number of potential points of failure for your service hosted on multiple servers, you can: