Maglev: The Load Balancer behind Google’s Infrastructure (Architecture Internals) — Part 2/3
In part 1, we gave an overview of what Maglev was about and why it can withstand significant scaling requirements given Google’s gargantuan size. In this article, we shall get more technical and look at the 3 main areas behind what makes Maglev a success. The technical aspects mentioned in the paper are quite dense, so I will use a few visualizations to elucidate the concepts
Once again, this article heavily borrows from the research paper “Maglev: A Fast and Reliable Software Network Load Balancer” released by Google. The link is at the end of this article.
1. Maglev Architecture
A Maglev instance is made of 3 main parts as shown in Figure 1 below. The Controller that houses the VIP announcer, the Config Objects and the Forwarder. We spoke about the Controller in detail in the previous article. Now we shall look at the Config Objects and mainly the Forwarder.
The Forwarder
At a high-level, the forwarder receives packets from the Network Interface Card (NIC), transforms the packets into an…
