Viasat Boosts Analytics Performance Using Amazon EBS

A Satellite Internet Innovator

The internet is increasingly at the center of modern life—from homes to businesses to government and the military—with 88 percent of U.S. adults connected as of 2016, according to the Pew Research Center. Yet not everybody has the luxury of a cable, fiber, or LTE connection. Viasat has the solution: high-tech, high-throughput communication satellites that provide fast internet access to customers on land, at sea, and in the sky. The company operates a broadband satellite ISP, provides Wi-Fi service for airline passengers and crew, and connects military and government operations securely and reliably to the data they need, no matter where they are.

Delivering these services requires a range of complex technologies on the ground as well as in the sky: hardware that converts electromagnetic waves into IP data, MAC layer processing, traffic accelerators, and more. These systems generate an average of 250 megabytes of data per second around the clock—data Viasat needs to understand in order to operate and continuously improve its services.

Unifying Data for Improved Insights

Collecting this data efficiently was challenging. The company gives its business units the freedom to operate with a large measure of independence. Although this enables agility and supports innovation, it led to many disparate, non-standardized data streams that the analytics team had to contend with. Additionally, the company uses a wide range of hardware, from custom-made, vertically integrated components to third-party networking equipment. “Managing the data-collection pipeline was a lot of work in this environment,” says Daniel Woodlins, software engineer at Viasat. “Information was coming in from many different types of devices and services, and much manual work went into standardizing it. Additionally, we had separate systems for monitoring and analytics data, which meant we weren’t always looking at the same information across the organization.”

The company also found that latency was an issue, particularly for a highly used business-intelligence system. “We had occasions where flushing data from cache to disk incurred latency,” says Woodlins. “That affected the responsiveness of time-series dashboards, which are the most highly used way to view data across the company.”

To solve these challenges, the team created a centralized system, known as Databus, for collecting, cleaning, and distributing data. Viasat selected Amazon Web Services (AWS) as the cloud platform for Databus after a positive experience using AWS to host a Hadoop analytics solution. Databus ingests, standardizes, and enriches incoming data and stores it on a managed Apache Kafka cluster running on Amazon Elastic Compute Cloud (Amazon EC2). In addition to adopting Kafka, the company moved from Amazon EC2 instance storage for main data stores to Amazon Elastic Block Store (Amazon EBS). This enables it to switch instance sizes on the fly without having to replicate data.

The solution writes 10–12 TB of data per day, and it stores 2 TB of that data after deduplication and standardization. Kafka brokers are hosted on an Amazon EC2 m4.2xlarge instance with three Amazon EBS st1 2TB storage volumes. The Kafka zookeeper and schema registry runs on an Amazon EC2 m4.large instance with an Amazon EBS gp2 100 GB storage volume.

With all the company’s data in one high-performance system, Viasat teams can easily use it for a variety of needs. The most visible to Viasat decision makers is the time-series database, which feeds monitoring and management dashboards. These allow Viasat to ensure systems are delivering the right levels of performance for customers.

Databus serves many other purposes as well. Security teams can analyze data to identify attacks, misconfigurations, and suspicious activity. Customer-care representatives can gain rapid insight into issues that may be affecting the user experience. The company also uses data for growth projection and capacity planning, making the best use of its satellite bandwidth. And Databus on AWS supports advanced analytics and experimentation to develop new technologies and approaches.

Optimizing Storage for Improved Cost and Performance

Viasat’s use of Amazon EBS st1 volumes—which are low-cost, hard-disk-drive (HDD) volumes designed for frequently accessed, throughput-intensive workloads—has been a major driver of technical and business benefits for the Databus project. “Adopting Amazon EBS st1 volumes smoothed out I/O performance for high-throughput data capture, improving the availability of our monitoring dashboards without increasing our storage cost.”

The team has also realized several other benefits. “Using Amazon EBS st1 volumes gave us the flexibility to choose instance sizes independent of I/O considerations,” says Woodlins. “Maintenance is faster, because we can reattach drives as needed instead of having to deploy and rebalance a new, empty server. We were also able to reduce the number of default replicas of our cluster from three to two because of the redundancy built into EBS volumes.” In addition, Viasat noted that using Amazon EBS st1 volumes enabled it to saturate instance I/O to the limit on the machine, resulting in maximum utilization and no need to overprovision.

The company plans to move more of its data storage to Amazon EBS st1 volumes. “Tests with our Hadoop workloads have shown great results,” says Woodlins. “Our long-term plan is to move them from Amazon EBS d2 instance store to st1 drives.” With the capability and flexibility of AWS Cloud storage, Viasat can continue to drive broadband innovation and performance ever higher.