Product description Fault tolerant load balancer

A fault-tolerant load balancer is a reverse proxy that distributes internet traffic between various Selectel services located in different regions and availability zones:

The load balancer can be used to build fault-tolerant services reserved in multiple regions and hybrid infrastructures based on dedicated servers, cloud servers, and Public Cloud powered by VMware. The load balancer also provides protection against DDoS attacks for all incoming traffic at the L3 and L4 levels.

There are no limits on the number of target servers or the volume of traffic transmitted.

Operating principle

To connect the load balancer, you combine your infrastructure into a private network via the Selectel Global Router and connect the load balancer to the same network. We provide you with an external IP address for the load balancer to receive traffic; this address is protected against DDoS attacks and is announced to the internet via BGP Anycast. Traffic incoming to the load balancer is distributed across the infrastructure behind it based on specified rules—load balancing algorithm, protocol, and the IP addresses and ports of the target servers.

The load balancer is reserved in two independent regions—Moscow and St. Petersburg; a pair of high-availability load balancers operates in both regions. If a router or load balancer in one region becomes unavailable, traffic is automatically switched to the other region.

Selectel is responsible for configuring the load balancer, ensuring fault tolerance, and monitoring.

Load balancing levels

According to the OSI model, balancing is possible at the following levels:

L4 over TCP (transport layer) — the load balancer redirects requests to servers according to specified algorithms;
L7 over HTTP (application layer) — the load balancer analyzes the request content and selects which server to send the request to.

Load balancing algorithms

Different load balancing algorithms are available at both levels (L4, L7):

Round Robin — a round-robin scheduling algorithm where requests are passed to each server in turn;
Weighted Round Robin — a weighted round-robin scheduling algorithm where each server is assigned a weight coefficient corresponding to its performance and capacity. Servers with a higher coefficient receive more requests;
Source IP hash — an algorithm where the preferred server to receive the request is chosen based on an HTTP header or IP address;
Least Connections — an algorithm where the request is sent to the least loaded server.

Comparison of load balancing levels

	L4	L7
Traffic type	Any	HTTP
Public ports	Any	443 with TLS (SSL) 80 without TLS (SSL)
Protocol	TCP	HTTP
SSL Offloading support	✗	✓
Let’s Encrypt support	✗	✓
Redirect from HTTP to HTTPS	✗	✓ (with active SSL Offloading)
HTTP/2 support	✗	✓ (with active SSL Offloading)
Source IP address forwarding	✓ (via proxy protocol if enabled on the backend)	✓ (via standard X-Real-IP header)
Active/standby support	✗	✗
Session support	✗	✓ (sticky sessions)

Limits

For a fault tolerant load balancer, limits are set on the number of concurrent connections. The limit depends on the selected pricing plan:

Start — 1 600 connections;
Basic — 4 000 connections;
Standard — 8 000 connections;
Ultra — 80 000 connections.