To Beat The Cloud, Datacenter Operators Must Embrace Automation, Multiple Vendors: Juniper VP

If datacenter operators want to compete with the cloud providers, they've got to start thinking more like them. That's what Mike Bushong, VP of Datacenter Business at Juniper, told The Register.

"These cloud principles simply have to creep into enterprise networking. Otherwise they're going to be outclassed as they compete against more and more of these SaaS-type companies," he warned.

This isn't an easy ask, he admits, because operators of datacenter networks don't have the luxury of starting fresh. "What you can do is claw back piece-by-piece some freedom architecturally by making good sound decisions over and over again."

Bushong's hypothesis is that datacenters need to start embracing the networking and infrastructure philosophies that made cloud providers successful.

"If you look at the most stark advances in operations, certainly it's led by the cloud companies," he observed. "The magic of cloud is not that the servers are in their datacenters instead of your datacenters. The magic of cloud – what you're really paying for – is operations. You're paying for the software teams that make the stuff work."

By contrast, most enterprise datacenters were built and networked to accommodate business-critical applications. The cloud providers flipped this model on its head, standardizing on infrastructure and operations, he explained.

What starts as a capacity problem quickly becomes an operations problem

When it comes to network automation, operators are spoiled for choice. Cisco has Application Centric Infrastructure (ACI), Arista has CloudVision, and Juniper has Apstra.

All of these platforms promise to simplify and automate network management, Bushong said – but none is perfect.

"A single pane of glass makes sense if you have a full ecosystem," he said, singling out Cisco. "I'm not sure that that's led to huge transformational benefits for customers."

The future is multi-vendor

In the past, datacenter operators have standardized on one or two networking vendors to minimize cost and complexity. But supply chain constraints are forcing many to reevaluate, because acquiring networking hardware has become harder amid global silicon shortages.

Those struggles have seen lead times for critical components stretch beyond a year. Last week, Arista announced plans to spend $4.3 billion this quarter to secure components, after supply chain constraints again overshadowed its earnings.

• Fortinet's latest firewall is like your kids' music – you're probably not ready for it, yet
• Intel sticks with FPGAs and ASICs for next-gen IPUs
• Arista Networks makes $4.3b bet on components amid supplies drought
• Was Nokia's Microsoft switch deal win a reward for SONiC contributions?

Bushong said he believes the future of the datacenter network will include multiple vendors – because doing so may be the only way to get the kit users need.

Unsurprisingly, Juniper's intent-based networking platform Apstra was built to handle multi-vendor networks.

Here again, there are some similarities to the cloud providers. Supply chain constraints drove Microsoft to adopt Nokia – a recent entrant to the datacenter switching market, Dell'Oro analyst Sameh Boujelbene recently told The Register.

According to Dell'Oro, Microsoft's multi-vendor philosophy has allowed it to maintain pricing pressure on networking vendors, while ensuring a steady supply of hardware.

Using Apstra, according to Bushong, operators can achieve similar results, without sacrificing manageability or complexity.

"Choice and flexibility – that's how they can get the right technology based on what they need. It's how they maintain economic leverage with their suppliers. It's how they reduce enterprise risk based on parts availability or component availability," he said.

SmartNICs and the edge are next

According to Bushong, the ability to support multi-vendor networks is only going to become more important as technologies like SmartNICs and the edge gain momentum.

SmartNICs – like Amazon Web Services' Nitro or Intel's suite of infrastructure processing units – are widely deployed by cloud providers to offload common infrastructure workloads from the CPU and free up resources for tenant workloads.

But achieving these benefits in enterprise datacenters isn't so easy, Bushong noted. "What starts as a capacity problem … actually quickly becomes an operations problem, because I get an explosion of endpoints."

Just like the rest of the network, every SmartNIC needs to be managed as part of a fabric, he explained.

To mitigate this challenge, Juniper has been working with Intel and Nvidia – another major SmartNIC vendor – to develop tools for managing the devices at scale via its Contrail platform.

Many of these same challenges also apply to the edge, Bushong added. "Large telcos are migrating from a small number of large national datacenters to a large number of highly distributed datacenters."

In doing so, those operators hope to achieve lower latencies at the expense of performance, but "the trade-off there is that instead of managing one central thing, I have to manage hundreds or thousands of distributed things." ®