Load Balancers

---

A load balancer can be deployed to distribute client requests across server nodes in a farm or pool.

• NIC teaming allows load balancing at the component level
• can use a load balancer in any situation where you have multiple servers providing the same function
  • E.g., web servers, email servers, web conferencing servers, and streaming media servers
• placed in front of the server network
  • distributes requests from the client network or Internet to the application servers
• service address is advertised to clients as a virtual server
• used to provision:
  • high availability services that can scale from light to heavy loads
  • fault tolerant services
  • provide mitigation against denial of service attacks

Two Types of Load Balancers

• Layer 4 switch
  • basic load balancers make forwarding decisions on IP address and TCP/UDP header values
  • works at the Transport layer of OSI model
• Layer 7 switch (content switch)
  • makes forwarding decisions based on application-level data
    • a request for a particular URL or data types like video or audio streaming
  • requires more complex logic

Info

• appliances can perform switch-like forwarding at layer 3, 4, and 7
• collectively referred to as multilayer switches

Scheduling

The scheduling algorithm is the code and metrics that determine which node is selected for processing each incoming request.

• simplest type is called round robin
  • means picking the next node
• Other methods include
  • picking the node with the fewest connections
  • or the best response time
• Each method can be weighted using
  • administrator-set preferences
  • dynamic load information
  • or both
• uses some type of heartbeat or health check probe
  • to verify whether each node is available and under load or not
  • Layer 4 load balancers can only make basic connectivity tests
  • layer 7 appliances can test the application’s state and verify host availability

Source IP Affinity and Session Persistence

Source IP or session affinity is a scheduling approach used by load balancers to route traffic to devices that have already established connections with the client in question.

• layer 4 approach to handling user sessions
• when a client establishes a session, it becomes stuck to the node that first accepted the request
• application-layer load balancer uses persistence to keep a client connected to a session
  • configuration option that enables a client to maintain a connection with a load-balanced server over the duration of the session
  • aka sticky sessions
  • works by setting a cookie either on the node or injected by the load balancer
  • can be more reliable than source IP affinity
    • but requires the browser to accept the cookie