Servers Heading Towards 100Gbps Connections

Research firm Dell'Oro Group has just published its Ethernet Controller and Adapter market global forecast. It predicts that in five years 44% of those shipments will support port speeds of 100Gbps or higher.

The rise of 100+ Gbps connections will be driven by the hyperscalers - AWS, Azure, Google, and social network operator Meta - ordering 100Gbps and 200Gbps SMARTNICs. These have far more processing capacity than traditional network interface cards.

SMARTNICs are likely to be adopted by other cloud providers, larger enterprises and firms in the telecoms industry.

What's Driving the Shift to 100Gbps?

As processors get faster and more power-efficient, more workloads and more demanding workloads can be hosted on a single high-spec server - increasing the aggregate bandwidth individual servers must accommodate.

Residential broadband is getting faster meaning the bottleneck in the end-user experience is often shifting from the broadband link to the server. For example, remote workers in the UK that upgrade from FTTC to FTTP could potentially see their download speeds increase 15x.

The shift to the cloud continues at pace, with workloads migrating to servers in purpose-built data centres where Internet bandwidth is comparatively cheap.

Ever larger amounts of data are being stored - creating demand for bandwidth as the data is saved, read, synced, backed-up, and scanned for malware.

Another driver is manageability. Once bandwidth usage gets beyond about 20Gbps, it's easier to use 100Gbps connections rather than lots of 10Gbps connections. By shifting to fewer, higher-bandwidth connections, flexibility and scalability are improved, and there are fewer network interface cards to go wrong.

But there's a problem. 100Gbps SMARTNICs are currently expensive and most networks aren't set up to accommodate 100Gbps connections. Those two impediments will lessen over the next five years.

What are SMARTNICs?

SMARTNICs are network interface cards with greater than usual capabilities. They make it possible for individual cloud servers to offload some of their networking and storage-related tasks to the SMARTNIC, freeing up CPU resources. This enables the servers to do more work. 

What Will Faster Server Connections Mean for Corporate IT

Faster server ports will help shrink backup windows significantly.

Hosted servers will get faster connections to the Internet - making websites appear more responsive.

High-capacity dedicated connections to public clouds will become more affordable, allowing firms to more easily combine Compute, object storage, virtual desktop infrastructure and offsite backup offerings from multiple cloud providers.

Workloads will continue to gravitate towards proper data centres where ultra-fast connectivity is cheap, plentiful and quick to upgrade. This will mean a shift towards cloud hosting and colocation.

We've already seen self-hosted corporate email servers being replaced with cloud-hosted subscription services such as Exchange Online. There's also been a shift from self-hosted PBXes to cloud-hosted alternatives. The connectivity speed advantages of cloud hosting will likely be just another minor factor nudging organisations towards replacing old on-premises servers with cloud-hosted alternatives once the old servers reach end-of-support.

What's After 100Gbps?

The Dell'Oro research confirms 200Gbps port speeds are already on the cards for 2027. One terabit per second will surely follow in the next five years. 

It's tempting to look at relatively high connection speeds and say "No one needs that much bandwidth. It will never take off." But history suggests otherwise.

In IT, supply often creates its own demand. How? By massively improving unit economics so new demand-creating services become economically viable (or cheap enough to be cross-subsidised for strategic reasons).

Network hardware vendors have revenue figures to maintain, so look to counter tech price deflation by upselling models with increased capacity, features and speed. Tech buyers know their organisations' bandwidth use is only likely to grow in the coming years, so many are perfectly content to take a future-proofing upgrade in lieu of a price cut that would only lead to capacity problems in a few years.

100Gbps Server Connectivity Shouldn't Be Viewed In Isolation

In the five years it will take for SMARTNICs to colonise the hyperscalers' data centres, Compute will get cheaper and faster. So will Internet access. Data storage will also get cheaper.

The aggregate impact of these changes will be more than the sum of their parts.

For example, we could see Virtual Desktop Infrastructure become economically viable as a mass-market proposition for businesses of all sizes. Microsoft is clearly betting on that, given its launch of Windows 365. 

Cloud gaming is likely to continue to grow in popularity. 

By the time 100Gbps is standard, we really will need faster server connections.

No doubt the hardware manufacturers are already planning what comes next.

Will hSo's Cloud Servers use 100Gbps?

Currently, our hosts do not need 100Gbps ports, as the customer workloads we host aren't that bandwidth hungry. Our own bandwidth-hungry workloads tend to be handled by task-optimised hardware (including application-specific integrated circuits) rather than our general Compute platform.

We expect 100Gbps SMARTNICs will fall in price significantly before we have a need for them, so there's no point in us buying prematurely. We'll wait for the hyperscalers to discover the bugs of bleeding-edge kit first.

We're always happy to invest in our platforms to stay ahead of rising demand. We are currently using smartNICs - just not 100Gbps ones yet.  We'll be doubling the Compute capacity of our cloud hosting platform over the coming months. It's likely to be several years until our hosts need 100 Gbps ports.

So, we'll sit on the sidelines, for now, waiting for the kit to get cheaper and more reliable. We suspect many corporate technology purchasers will do the same.