PARIS -- MPLS, SDN and NFV World Congress 2018 -- Internet giant Google is developing a "Google Assistant for Networking" tool that makes use of intent-based networking technologies to cope with the growing complexity of its architecture.
The assistant is designed to automate some of the processes used in managing data centers and other infrastructure, and shows that Google faces similar challenges to telecom operators as data traffic soars.
Vijoy Pandey, Google's head of engineering for data centers and backbone networks, says that about a quarter of Internet traffic now originates from Google (Nasdaq: GOOG) and that "stuff breaks at scale."
In a presentation at today's MPLS, NFV and SDN World Congress in Paris, he revealed that Google has about 10,000 switches in operation in a typical large data center and is now handling around 30,000 configuration changes every month. Its search engine currently processes about 3.5 billion searches every day.
"What really stresses out the network is humans," he told his audience at a keynote presentation in Paris. "It is actually humanly impossible to do changes in the size of the network we have. Humans are looking at the design elements but software needs to handle the nitty-gritty."
Pandey estimates that about 70% of failures happen when a management operation is in progress. The apparent aim is to make improvements through even greater reliance on automation.
Google is using various data models to build the assistant, according to Pandey. These include a Google-specific model for network topology, called the unified network model (UNM), as well as OpenConfig for configuration and telemetry. Domain specific languages (DSLs) are being used in the policy area.
"We want to capture everything in the data model that automation needs," said Pandey. "We don't want data in people's heads or in spreadsheets. That is not automation. You need a detailed data model."
What remains unclear is the current status of Google's investment in this area. Pandey was evasive when asked about this during a Q&A session with audience members. "UNM is deployed and that all comes out of necessity," he said. "Where we are going is a bit harder to answer but think of all this as what we are doing in the design space."
The UMN, he explained, covers the physical, transport and packet layers and includes metadata for network planning, automation and simulations.
In comparison with telecom operators, Google already relies on an extraordinarily small team of people to manage its data centers and other infrastructure.
Virtyt Koshi, the EMEA general manager for virtualization vendor Mavenir, reckons Google is able to run all of its data centers in the Asia-Pacific with only about 30 people, and that a telco would typically have about 3,000 employees to manage infrastructure across the same area. "The economics of it are incomparable," he says.
With about 80,000 employees in total last year, Google parent Alphabet was generating roughly $1.4 million in revenues for each employee. The equivalent figure for Verizon, one of the world's most efficient operators on this measure, was about $810,000.
Interest in the use of intent-based networks and other automation tools has fueled concern about the prospect of looming job cuts in the telecom sector. In a recent Light Reading survey, cutting jobs emerged as the number-one reason for automating network operations. (See Automation Is About Job Cuts – LR Poll.)
Some of the world's biggest operators have already slashed thousands of jobs in the last couple of years, although the axe appears to have fallen heaviest on customer services and sales staff at many companies. (See Efficiency Drive by Major Telcos Has Claimed 74K Jobs Since 2015.)
— Iain Morris, News Editor, Light Reading