Industry veteran relates the fun and challenge of building the world-class Oracle Bare Metal Cloud network.
By Alexandra Weber Morales
When you’re “an engineer who builds stuff,” like Matteo Frigo, life’s not just about perks and stock options. A lifetime of algorithmic expertise earns you access to the best people—and the hardest problems.
Matteo Frigo, Oracle Software Architect
Over his career, Frigo has applied his expertise to a range of problems: developing medical breath analyzers, authoring a popular library for Fourier transforms, and pioneering the Cilk language and runtime system for parallel programming. He first made his name helping invent cache-oblivious algorithms, meaning algorithms that run optimally without hardware-specific tuning.
After architecting the Amazon Web Services (AWS) Elastic File System a year and a half ago, Frigo wanted a new challenge. When many of his former AWS colleagues began populating Oracle’s cloud team in Seattle, Frigo’s curiosity was piqued: “At some point, I decided Amazon wasn’t for me. Then Oracle said, ‘You can be the Boston branch of Oracle Bare Metal Cloud.’”
Now an Oracle software architect, Frigo is applying his deep knowledge and experience—and fielding a team of Boston-area technologists—to build a data center network that makes the Oracle cloud infrastructure service unique.
Building a virtualized network that is 4 to 10 times faster than today’s highest-performing networks is technically demanding—and Frigo’s team relishes that class of problem. “We’re looking at 10 nanoseconds to process every packet that comes in,” he says. “If you take that requirement seriously, you need a lot of engineering to get to that point. It’s not something that you can get to incrementally. You have to get it right from the beginning.”
Building it right means having the right team, which started with just Frigo in January 2016 and is now up to 15 people, with plans to add another 15 by June 2017. As the technical lead, Frigo defines the project and the architecture, “but don’t think of me as someone who just writes design docs. I spend most of my days writing code, and the most difficult parts of the code,” he says.
The main challenge? Create what Oracle Executive Chairman and CTO Larry Ellison has called a second-generation cloud offering. Oracle Bare Metal Cloud Services emulate on-premises data centers for full operational control, maximum isolation, and high performance, while still providing all the scalability and efficiency of a cloud. At the core of the offering is a virtual network that is much faster than customers can make themselves, Frigo says. “Giving customers the feeling that they have a network to themselves requires some secret sauce,” he adds.
Part of Oracle’s approach is to let technologists like Frigo apply their experience and perspective to the right problems. Here is some of that perspective.
“Most people think shipping quickly means you build something that may not be complete but provides important features,” Frigo says. “I think it means you think hard to implement a simple solution to the right problem. Many will try to solve the immediate need and not think beyond that.”
Don’t think of me as someone who just writes design docs. I spend most of my days writing code.
For example, “When a given network packet comes in, should you forward it or not?” he says. “The quick implementation is to examine only the IP addresses where the packet comes from and where it is going, but that’s not what we did. We built a general-purpose pattern-matching engine so we can compute any property of the packet that we care about. Implementing the general solution takes roughly the same time as implementing the specific solution, but now I have a system that can cope with the inevitable requirement changes without extra development time on our part. In the end, you ship faster if you focus not on the specific use case but on a more general situation that distills the essence of the problem that you are trying to solve.”
“The thing that makes it fun is the interaction with brilliant people on a day-to-day basis,” Frigo says. “Half of them are PhDs from MIT. It’s a fun environment, and we’re constantly getting things done at a much faster pace than any other team I’ve worked with. There are companies that try to force things like project tracking and other nonsense that is just overhead. We do something we like, and we don’t need any external motivation.”
Frigo says Agile methods can be useful, depending on the situation. “If you are doing consulting work and the requirements aren’t clear, it makes sense to do a quick prototype,” he says. “That works for problems where the hard part is trying to figure out what to build. But there are other types of problems where the thing is hard to build. If you have to deliver something that looks like an airplane every three weeks, Agile won’t work. In Oracle Bare Metal Cloud, we have both kinds of work.”
“The first reason is that the company is that serious about Oracle Bare Metal Cloud. It’s clearly strategic, and there’s lots of support from Larry Ellison and the board. Oracle understands that building this right is crucial,” Frigo says. “Another reason is that most of the people who are architecting this cloud at Oracle have already built Amazon, Azure, or Google. We’ve all already made a ton of mistakes and are eager not to repeat them.”
As a result of Ellison’s focus, “we have a lot of freedom to do the right thing and build it right,” he says. “Having a directive from the top that orients the whole company in the direction of the cloud is a major reason to be at Oracle and not somewhere else.”
This article was originally published in Oracle Magazine.
Alexandra Weber Morales, principal at World Wind Writing, is the former editor in chief of Software Development magazine and has more than 15 years of experience as a technology content strategist and journalist.