Arm in the Datacenter: What to Expect Next
Today, the question isn’t “does it work” — it’s all…
ScyllaDB is all about pushing limits.
As a drop-in alternative to Apache Cassandra and Amazon DynamoDB, the real-time NoSQL database company aims to power applications with ultra-low latency and extreme throughput. Their ultimate goal? To unleash infrastructure’s true potential by extracting maximum performance from modern hardware.
“We deal with massive amounts of either data, transactions, or both for customers in many verticals, such as AdTech, IoT, and customer retail,” says VP of Field Engineering Glauber Costa. “We run everywhere. We run on virtualized environments, but we also run on bare metal machines—and that’s where you get the most out of our architecture.”
As a component of the company’s continuous integration pipeline, “we use the Arm machine to verify that everything builds, runs, and works,” says ScyllaDB Co-Founder and CTO Avi Kivity. “Through the Works on Arm initiative our friends at Packet gave us a machine free of charge to do testing and development, and we still use it to this day.”
Last year, Costa and his team got an early preview of the Graviton2-based chips. Without a doubt, Costa predicted these next-generation Arm-based processors would bring “seismic” changes to the cloud-native ecosystem.
“We’re seeing new Arm processors that are going to shake up the industry,” says Costa.
Traditionally when comparing Arm to x86, Arm had the upper hand in terms of affordability. That reputation, however, came at a cost. “In the past, we assumed from the get-go that because an Arm processor costs less, the performance level would suffer,” Costa says. “There was a trade-off imposed on users, and it always felt like you were leaving something on the table with Arm.”
Not anymore. When Costa set out with ScyllaDB Software Engineer Vladislav Zolotarov to benchmark the new AWS Arm-based servers, “we were very surprised to see that they still cost less, but they were not slow at all,” Costa says. “In fact, some workloads were faster compared to x86.”
Add to that other well-known attributes of Arm instances such as energy efficiency and ubiquity in the IoT and embedded world, “it is impossible to ignore Arm-based systems and the benefits they offer,” Costa says.
As Arm instances gain popularity, the ScyllaDB team has taken steps to ensure that the transition for its business partners will be seamless.
“We are prepared for it,” Kivity says. “For us, it will be easy to go from x86 to Arm since everything is portable and already part of our testing pipeline.”
So if current ScyllaDB users—which include Comcast, IBM, and Starbucks—or new ones inquire about Arm, “we will be able to satisfy their needs very quickly,” Costa says.
There has always been “a great deal of synergy” between ScyllaDB and Packet, says Costa. Last year, when ScyllaDB set out to benchmark the power of its close-to-metal architecture, they leveraged Packet’s bare-metal cloud.
“We wanted to push the limit and we scaled to one billion rows per second,” Costa says. (Yes, he said “billion!”) Not only does that mean Scylla delivers faster applications and analytics, but by leveraging bare-metal, “the cost of scaling out is greatly reduced,” Costa says.
Now ScyllaDB is looking ahead to offer its scale-out, scale-up database to customers both on-premise and also on various public clouds, including its own cloud offering Scylla Cloud, which runs on top of Amazon.
“We are trying to expand by increasing our value proposition and increasing our compatibility with more databases such as Cassandra or DynamoDB,” Costa says. “We want to help customers reduce the cost of running their workloads and deliver more reliable service so they, in turn, deliver better SLA.”
And they believe Arm will help them do that in the long run because, Costa says, “that’s a platform that we will always bet on.”