Teridion uses BigQuery and Cloud Machine Learning Engine to provide SaaS applications with a high-performance, highly reliable virtual network wherever they are in the world.
Google Cloud Results
- Predicts optimal network routes with a 99% accuracy with Cloud Machine Learning Engine
- Loads data up to 20 times faster than the public internet
Routes hundreds of terabytes of data every day
Teridion provides Software-as-a-Service companies with high-speed, highly available virtual networks that dynamically route traffic through the best-performing Internet paths. “We use around 20 different cloud services to try and eliminate geographical borders for SaaS applications,” says Elad Rave, Founder and CTO at Teridion. “Our customers expect to get the same high performance wherever they are in Asia Pacific, Europe, the Middle East, Africa, or the Americas.”
“We signed some big customers in our first year and we started handling very significant amounts of data. We had to move to a more scalable solution quickly and after some trials and cost/performance analyses, it was clear that Google was the best fit for us.”—Elad Rave, Founder and CTO, Teridion
Teridion launched in 2014, born of Elad’s frustration trying to solve network problems with a previous startup. After experimenting with content delivery networks (CDNs), he hit upon the idea of changing traffic paths through different cloud companies to avoid outages and slowdowns, and formed Teridion. Within a year the company was routing so much traffic that it needed a new, cloud-based infrastructure to keep up with its clients’ needs. To do that, it turned to Google Cloud Platform (GCP).
“We signed some big customers in our first year and we started handling very significant amounts of data,” says Elad. “We had to move to a more scalable solution quickly and after some trials and cost/performance analyses, it was clear that Google was the best fit for us.”
Identify the problem, find the solution
To provide the best possible service to the company’s clients, Teridion does two things. “First, we look for where the problems in the internet are, and then we can route traffic around it,” says Elad. The heart of the company is the Teridion Management System (TMS), which analyzes client traffic data and computes a dynamic path that changes according to the circumstances. By early 2015, Teridion needed an upgraded infrastructure to deal with the huge amounts of data it was handling. After evaluating the leading cloud providers, Teridion chose Google as the new core for TMS.
“With Cloud Machine Learning Engine, not only do we know where the problems are right now, we can also predict where they’re going to be a few hours from now. That has been very effective for us.”—Elad Rave, Founder and CTO, Teridion
The Google environment was already familiar because Compute Engine was a major component of Teridion’s routing service. The company was also a longtime user of G Suite which helped implement a cloud-based mindset. When it came time to migrate, Teridion planned everything in advance and executed “within just a couple of weeks”, says Elad. The first big change was moving the Teridion Management System to BigQuery. This allowed TMS to become a highly scalable streaming database without the headaches of constant maintenance.
“BigQuery was the first external service we used for our database and it allowed us a much greater level of automation,” says Elad. “The more we used it, the more capabilities we could add. That’s when we started looking at machine learning and artificial intelligence.”
Cloud Machine Learning Engine (Cloud ML Engine) provided Teridion with the opportunity to transform its routing service from one that detected and reacted to problems in the network, to one that predicted and avoided the problems altogether. Teridion used machine learning tools from Google to integrate seamlessly with BigQuery and learn from the data being collected, leading to prediction models that could be fed back into the routing service.
“With Cloud Machine Learning Engine, not only do we know where the problems are right now, we can also predict where they’re going to be a few hours from now. That has been very effective for us,” says Elad.
“Our clients get great performance, which keeps their users engaged and helps reduce the cost of customer service. What we’ve been able to achieve with machine learning is really exciting.”—Elad Rave, Founder and CTO, Teridion
Highly scalable, highly available
GCP has helped Teridion to dramatically expand its business over the past few years without compromising on performance for the company’s clients. “We’re handling several hundred terabytes a day, creeping up to petabytes,” says Manav Mishra, Vice President of Products at Teridion. “As well as operating at a huge scale, Teridion provides its customers with a network that accelerates traffic up to twenty times faster than the public internet and is significantly more available. Our machine learning algorithms are able to predict optimal routes with more than 99% accuracy, which means that clients rarely, if ever, suffer outages. In 2017, TMS detected a major Border Gateway Protocol hijacking within minutes and rerouted the networks to safety. However, our clients maintained 100% availability during a global outage event.”
“Our clients get great performance, which keeps their users engaged and helps reduce the cost of customer service,” says Elad. “What we’ve been able to achieve with machine learning is really exciting.”
Continuous integration, continuous delivery with Kubernetes
The next milestone for Teridion is to transform its infrastructure to run on a fully automated Kubernetes-based architecture, orchestrated with Kubernetes Engine. The demands of constantly updating the machine learning components of Teridion’s system led the company to prototype a serverless solution. “What Kubernetes will give us is a highly scalable, easy to deploy, and highly available service that can be maintained and updated in a pure continuous integration/continuous delivery environment,” says Manav. With a prototype just completed, Teridion expects to roll out the new architecture later in 2018. In the meantime, Teridion will keep working to find new ways to deliver the best possible service for its clients.
“We’re trying to do as much as we can with Google,” says Elad. “We push a lot of data, but so far Google Cloud Platform has been able to keep up with us.”
Teridion’s Internet Overlay Network makes SaaS applications faster and more reliable by improving Internet performance up to 15x, anywhere in the world.