Google this week did something it rarely does: It revealed details of some of its networking technology.
Specifically, it shared information about its software network load balancer — Maglev — that enables Google Compute Engine load balancing to “serve a million requests per second with no pre-warming,” according to its corporate blog posting.
“Google has a long history of building our own networking gear, and perhaps unsurprisingly, we build our own network load balancers as well, which have been handling most of the traffic to Google services since 2008,” says the blog. “Maglev load balancers run on ordinary servers — the same hardware that the services themselves use.”
Essentially, Google created its own virtualized network function (VNF) load balancer several years ago.
Google says hardware load balancers waste half of their capacity running in active-passive configuration to provide failover. Its Maglev software load balancers spread incoming packets across Maglev instances, “which then use consistent hashing techniques to forward packets to the correct service backend servers…
“With Maglev, it’s easy to add or remove load balancing capacity, since Maglev is simply another way to use the same servers that are already in the cluster.”
Other Shy Giants
Google provided the details about Maglev at the Usenix Symposium on Networked Systems Design and Implementation this week. Just last week, Google made its debut appearance at the Open Compute Project (OCP).
At OCP, which was created by Facebook, Google seemed to have a very specific agenda: getting all webscale companies on the same page in terms of data center power specifications.
It behooves these giant Internet companies to agree on certain specifications, says Dell’Oro Group analyst Alan Weckel. “There has to be some consistency for the supply chain to survive,” he says. “You need the entire industry behind certain things to get economies of scale. They all have an interest in getting some common components.”
Microsoft also was at OCP, garnering the spotlight with its Software for Open Networking in the Cloud (SONiC), which is open source software to run cloud switches.
“Microsoft has been really honest about what they share and what they have in production,” says Weckel.
At OCP there was one conspicuously absent webscale giant: Amazon Web Services.
Interesting piece. What’s great is that anyone can net the same benefits as Google without spinning up developers to get it done. The Zeus Extensible Traffic Manager (ZXTM) was the first widely deployed software load balancer that I am aware of, brought by Damian Reeves (now of Google) and Adam Twiss around 2002 if memory serves. The ZXTM evolved from the Zeus Load Balancer software developed for their then leading-performance web server. The highly scriptable, high performance LB is still a leading solution today, as part of a Brocades NFV portfolio. It’s now called the vADC and has been optimized with Intels DPDK for scale out performance.