Learning F5 BIG-IP on a real data-centre fabric running on SONiC switches (part 5): load-balancing methods, measured

Part 5. Static and dynamic load-balancing methods on F5 BIG-IP: round robin, ratio, least connections, fastest, observed and predictive, how they react to a slow server, and how to choose and test one.
Hand-drawn spine-leaf fabric with a firewall, a BIG-IP and an app: part 5, load balancing measured

On this page

This is part 5 of a series on learning F5 BIG-IP Local Traffic Manager on a real data-centre fabric running on SONiC switches. Each part goes with a video on the LevelUP I.T. YouTube channel. Part 3 built the first virtual server, and part 4 took SNAT apart. This part measures the load-balancing methods.

Same pool, same three servers, the same 30 downloads, and one server deliberately slow. Round robin still gave the slow server its full share: 11, 10, 9. Least connections noticed and sent it fewer: 13, 10, 7. Nothing else changed. This part measures round robin, ratio, least connections and fastest on the lab, counted on the BIG-IP itself, and explains observed, predictive and the pool settings around the method.

Round robin vs least connections with a slow server3: 11/10/9 vs 13/10/7
Round robin vs least connections with a slow server3: 11/10/9 vs 13/10/7

What you'll see

  • Static vs dynamic methods, and member vs node.
  • Round robin, 99 new connections: 33/33/33. Ratio 3:2:1 over 120: 62/39/19.
  • A server made slow with tc netem: about 7 seconds per 4 MiB download, against under one second for the others.
  • Round robin, least connections and fastest with that slow server, with open connections watched second by second.
  • Observed and predictive, weighted and ratio least connections, slow ramp time, priority groups, limits and service-down action.
  • How to choose a method, and how to test one so the numbers mean something.

Static vs dynamic

Every method answers one question: for this new connection, which pool member? Static methods answer from the configuration alone. Round robin takes the next member; ratio does the same with weights you set. Dynamic methods answer from live state. Least connections picks the member with the fewest open connections; fastest picks the quickest responder; observed and predictive turn connection counts over time into a ratio.

Most methods come in a member and a node version. A member method counts only inside this pool. A node method counts the server across every pool it's in. Every method decides once per new server-side connection, after the client side is up and, with an HTTP profile, after the request is read. Requests on a kept-alive connection follow it.

Static and dynamic methods
Static and dynamic methods

This post is for paying subscribers only

Subscribe to LevelUp I.T. newsletter and stay updated.

Don't miss anything. Get all the latest posts delivered straight to your inbox. It's free!
Great! Check your inbox and click the link to confirm your subscription.
Error! Please enter a valid email address!