Packet Says Your Data Center Shouldn't Have an Opinion

Christine Hall

July 26, 2017

5 Min Read
Microsoft employee demonstrates the Microsoft HoloLens augmented reality (AR) viewer in March 2016 in San Francisco
Microsoft employee demonstrates the Microsoft HoloLens augmented reality (AR) viewer in March 2016 in San FranciscoJustin Sullivan/Getty Images

Packet, the New York-based startup that provides dedicated high-performance bare-metal servers as a cloud service similar to AWS or Azure, has added 11 data center locations around the world to its previously existing four and launched a new edge computing service aimed at applications that require low-latency communication between end devices and compute infrastructure.

The new service is called Edge Compute, and the reason for such rapid expansion is to make the company’s services more attractive to companies with those latency-sensitive workloads. But in general, Packet — which last year raised a $9.4 million Series A round led by the Japanese telco and technology investment powerhouse SoftBank — is on a mission to give developers access to unopinionated infrastructure. We’ll explain what that means later in the article.

“It’s no longer the case that having infrastructure in one or two locations is reasonable,” Zachary Smith, Packet’s co-founder and CEO, explained to Data Center Knowledge. “Having infrastructure in 20 locations might be table stakes for a latency-focused application. We consider this a chasm that’s been created between the people who can do this and those who can’t.”

See also: Vapor IO to Sell Data Center Colocation Services at Cell Towers

The locations that became available today are Los Angeles, Seattle, Dallas, Chicago, Ashburn (Virginia), Atlanta, Toronto, Frankfurt, Singapore, Hong Kong, and Sydney. This is in addition to the company’s existing locations in New York, Sunnyvale, Amsterdam, and Tokyo. Another expansion is scheduled in October, which will add Paris, London, São Paulo, and Mumbai to the mix.

“What we’re doing is taking our platform that’s already in four locations globally and expanding it to within five or ten milliseconds of all the major population centers,” he said.

Edge Computing for IoT, Self-Driving Cars, and More

As the name suggests, the new edge computing service is about better serving the outer perimeter of the internet, which is expected to see rapid growth with 5G, the upcoming high-speed mobile network standard expected to enable everything from more sophisticated — and-bandwidth hungry — Internet of Things devices to self-driving cars and augmented and virtual reality systems.

“While edge compute is still in its infancy, new experiences are driving demand for distributed infrastructure, especially as software continues its relentless pursuit down the stack,” Smith said. “We believe that the developers building these new experiences are hungry for a distributed, unopinionated, and yet fully automated compute infrastructure, and that’s what we’re bringing to the market today.”

For the time being, the new Edge Compute locations feature a single server configuration, Type 1E, which features an Intel E3-1578L v5 processor, Intel IRIS GPU, 32GB of RAM, 240GB of SSD storage, and 10 Gbps network interfaces. Packet is planning to add more configurations later, including servers powered by ARMv8 processors. The Type 1E lists at $0.50 an hour, but is also available through Packet’s new spot market, similar to AWS’s Spot Instances, offering marketplace-style pricing in all of Packet’s 15 locations.

For users requiring customized hardware, Packet has a program it calls Private Deployment which has been extended to the new data center locations. Through this program, users can deploy custom configurations while benefiting from Packet’s platform and network automation.

Equinix and Interxion

For the record, in North America and Asia Pacific, Packet generally houses its servers in data centers operated by Equinix, and in Europe it partners primarily with Interxion. Smith said the amount of floor space the company occupies varies from data center to data center.

“In some of our major facilities we’re obviously buying wholesale by the megawatt, because we’re in the cloud business and have thousands of servers. Some of the edge locations are fairly primitive, starting with small cages and allowing us to expand from there.”

Many Locations, No Opinions

Packet’s unique model helps remove infrastructure restrictions and the layers of abstraction that are inherent in most public clouds, Smith explained. “For any sort of innovative startup, it’s almost an impossibility for you to deploy infrastructure at 15 or 20 locations around the globe [on your own]. You’ve got to rely on somebody else’s infrastructure. At that point, you’re really consuming their opinion, and their viewpoint, and their abstraction away from the hardware.”

Indeed, Packet isn’t your father’s cloud company.

Unlike AWS, Azure, and GCP, which are based to a great extent on proprietary technologies and come with a degree of vendor lock-in, Packet is focused on the hardware. Software is available — plenty of operating systems, OpenStack, and more — but users can take the BYO approach and bring their own software — including operating systems. All instances are spun up on bare metal, with customers getting a fully isolated dedicated server with no shared resources. Packet supplies an extensive set of automation tools, but otherwise hands customers the cloud equivalent of an on-prem server, without vendor lock-in.

“At its simplest, we give you a physical server as if you went into a data center and racked it,” Smith said.

This means it’s not for everybody. It’s designed for experienced DevOps folks. Those who want to deploy containers with a couple of clicks without really knowing what they’re doing should probably look for another solution. According to Smith, about a third of the workloads on Packet are cloud native, specifically Kubernetes, Docker, Mesosphere, or automated DevOps services. Another large portion is enterprise IT, and includes applications running on OpenStack, VMware, CloudStack, and Hadoop.

“We offer, effectively, primitives,” he said. “The ability to automate hardware, the ability to choose your own operating system (or build your own), or the ability to choose your own networking stack. We’re doing so in a way that’s very developer-friendly, so you can use DevOps tools such as Ansible, Terraform, Libcloud, jclouds, or whatever you want — or your own scripting — to automate that hardware as if it was in your own data center.”

That’s exactly the way Smith and his team want to keep it.

“We have no visions of moving up the stack to do a proprietary or opinionated service. There are enough options in the marketplace for that.”

About the Author(s)

Christine Hall

Freelance author

Christine Hall has been a journalist since 1971. In 2001 she began writing a weekly consumer computer column and began covering IT full time in 2002, focusing on Linux and open source software. Since 2010 she's published and edited the website FOSS Force. Follow her on Twitter: @BrideOfLinux.

Subscribe to the Data Center Knowledge Newsletter
Get analysis and expert insight on the latest in data center business and technology delivered to your inbox daily.

You May Also Like