Google recently revealed details of Maglev, its network load balancing technology that gives developers the ability to build infrastructures that can handle a million requests per second, without warning. Maglev leverages Equal-Cost Multi-Path routing technology to disperse incoming network packets across all Maglevs in conjunction with “consistent hashing techniques” that enable the accurate transmission of packets to the “correct service backend servers,” regardless of which Maglev receives a specific packet. Maglev’s use of Equal Cost Multi-Path routing technology differentiates from the common industry practice of using Active Passive load balancer configuration, wherein the secondary, or passive load balancer configuration operates passively and awaits the opportunity to assume responsibility for load balancing as required. Whereas Active Passive load balancer configurations can waste half of their load balancing resources, Maglev’s use of ECMP enables a more deeply engaged utilization of existing resources as noted in Google’s blog post, below:
All Maglevs in a cluster are active, performing useful work. Should one Maglev become unavailable, the other Maglevs can carry the extra traffic. This N+1 redundancy is more cost effective than the active-passive configuration of traditional hardware load balancers, because fewer resources are intentionally sitting idle at all times.
Borg, Google’s cluster management technology, renders it possible to migrate service workloads between different clusters as required. Similarly, Maglev facilitates the addition and removal of load balancing capacity and thereby illustrates the capability of Network Functional Virtualization technology to add and remove load balancing functionality without the addition of new hardware. Google’s deep dive into the workings of Maglev, which runs on commodity Linux servers, illustrates how its technology manages load balancing at scale and facilitates its management both of Google network traffic more generally as well as load balancing within the Google Cloud Platform. Named after the Japanese bullet qua magnetic levitation train, Google’s Maglev technology requires the addition of new Maglevs once a certain threshold has been reached with respect to the use of existing Maglevs for network load balancing purposes.