F5 BIG-IP on a SONiC EVPN fabric, built live (Fabric Live 3): load-balancing methods and cookie persistence

Fabric Live 3: BIG-IP load-balancing methods, static and dynamic, tested against a server that is up but struggling, then every kind of cookie persistence (insert, rewrite, passive, hash) with the cookie decoded, and source address persistence as the contrast.
Hand-drawn spine-leaf fabric with a firewall, a BIG-IP and an app: Fabric Live 3, load-balancing methods and cookie persistence

On this page

This is part 3 of Fabric Live, a series that builds an F5 BIG-IP lab on a SONiC EVPN fabric and configures it live on camera. Each part goes with a video on the LevelUP I.T. YouTube channel. The lab itself, and how it is built (one file of values, one up.sh, the SONiC EVPN fabric, the Palo Alto firewall, the three BIG-IPs and the validation that proves it all works), is the subject of the first video: Fabric Live 1. Part 2 built the three deployment designs on that fabric: two-armed, one-armed and nPath. This part takes the two-armed design and goes deep on two things every BIG-IP deployment depends on: how the BIG-IP chooses a pool member, and how it sends a client back to the same member afterwards.

In this part I configure the load-balancing methods that matter in practice, static and dynamic. I make one server busy on purpose to show the difference, and then build every kind of cookie persistence, with the cookie decoded byte by byte.

What you'll learn

  • What a load-balancing method decides, which members it chooses from, and why a BIG-IP with eight TMMs never splits a small test evenly.
  • Why the BIG-IP balances connections, not requests, and what OneConnect changes about that.
  • Round robin, ratio, slow ramp, and priority group activation with a sorry server, including the window between a failure and its detection.
  • Least connections, fastest, weighted least connections, and observed and predictive, tested against a server that is up but struggling. Observed and predictive follow how the load changes, not how high it is.
  • The order in which persistence and load balancing run on every new connection.
  • Cookie insert, with the cookie decoded live, and what happens when the persisted member dies.
  • Cookie rewrite and its BIGipCookie placeholder, cookie passive and cookie hash.
  • Source address persistence, its records, timeout and mask, and why none of the cookie methods can work on the nPath design.

What a load-balancing method decides

A load-balancing method answers one question for every new connection: which pool member gets it? It is a property of the pool, not of the virtual server: in the web UI it sits on the pool's Members tab, in tmsh it is the pool's load-balancing-mode. The BIG-IP offers many methods, and they fall into two families.

Static methods decide from configuration alone: the order of the members, or weights you assign. They never look at the servers. Round robin and ratio are static. Dynamic methods decide from what the BIG-IP measures: how many connections each member has open, how many requests it has not yet answered, how those numbers change over time. Least connections, fastest, weighted least connections, observed, predictive and dynamic ratio are dynamic. Static is predictable and cheap. Dynamic adapts when the servers are unequal, or when one of them gets into trouble.

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!