There’s a lot of buzz in the industry right now about next-generation 400 Gigabit speeds that are being adopted by Tier 1 hyperscale data center providers like Google, Amazon, and Microsoft. Tier 2 and 3 cloud service providers are expected to ramp up to these speeds next year with large enterprise likely starting to follow suit in 2023 and 2024.

While 400 Gigabit speeds will eventually make their way into large enterprises for uplinks between switch tiers to handle increasing amounts of data, server connections are where bandwidth and latency need to keep up with e-commerce and emerging technologies like advanced data analytics, machine learning, artificial intelligence (AI), telemedicine, online banking, high-resolution video content, and other real-time applications. Thankfully high-speed interconnect direct attach cables (DACs) are keeping up with increasing requirements, ensuring that switch-to-server connections don’t become the weakest link.

They Get You Beyond 10 Gig

Most enterprise data centers employ server connection speeds of 1 or 10 Gigabit per second (Gb/s) using copper-based cabling with 10, 40, or 100 Gigabit fiber uplinks between switch tiers, with larger enterprises primarily running 10 Gig server speeds. 10 Gb/s server connections are achieved either using 10GBASE-T with category 6A structured cabling that supports up to 100 meter lengths or SFP+ DACs in direct short-reach connections from top-of-rack (ToR) switches where lengths are less than 7 meters.

While not an option for data centers that need to support longer distances, SFP+ DACs in a ToR deployment have become increasingly popular due to requiring less power per port and offering lower latency than 10GBASE-T. ToR switches with DACs typically use a lot less than 1 W per port, while 10GBASE-T switches range from 1.5 to 4 W per port. Latency with ToR and DACs is around 0.3 microseconds per links, while 10GBASE-T with its more complex encoding schemes is closer to 3 microseconds per link. A couple of microseconds might not seem like much, but emerging applications like high-speed trading and AI are increasingly demanding sub-microsecond latency. This makes DACs ideal for any current or future application where latency is a concern and where high port counts can add up to significant power savings. DACs are also easy to deploy-as a factory-tested and terminated solution they can be simply plugged in without the complexity of cable testing and multiple connection points.

As emerging technologies demand more bandwidth and lower latency, large enterprise data centers are now beginning to adopt 25 Gb/s server connection speeds. In fact, a recent five-year forecast report from Dell’Oro Group predicts that 25 Gb/s will gradually replace 10 Gbps for server speeds over the next five years. While 25 Gb/s can be supported using transceivers and duplex fiber connectivity (i.e., 25GBASE-SR), this is the most expensive option that is really only needed for very long switch-to-server lengths, which are extremely rare in enterprise data centers. Longer lengths that warrant transceivers and fiber cabling in the enterprise are typically only found in switch-to-switch links.

SFP technology has thankfully kept up with the need. SFP28 DACs that use the same form factor as SFP+ DACs support 25 Gb/s server connections-and the benefits of reduced power consumption, lower latency, and lower cost hold true at these speeds. In a comparison of power consumption for 500 server connections using SFP28 DACs verses 25GBASE-SR, the total wattage estimated with SFP28s is just 25 W vs. around 600 W for 25GBASE-SR.

At higher 25 Gb/s speeds, SPF28 passive DACs are however limited to about 5 meter lengths. While this still supports in-cabinet ToR switch-to-server deployments, data centers needing longer lengths can also look at SFP28 AOCs as an alternative to transceivers and fiber cabling. Able to support up to 100 meters and typically used for 30-meter link lengths and below, AOCs offer the benefit of less cost and power consumption than using transceivers with fiber cabling while offering smaller-diameter fiber cabling. For more information on the difference between DACs, AOCs, and fiber cabling with transceivers, check out our previous blog.

DACs Remain a Viable Option for Years to Come

The next logical migration in enterprise server connections will be 50 Gb/s, which will likely start to take hold as fiber switch-to-switch links increase to 200 and 400 Gb/s over the next 3 to 5 years. The good news is that emerging SFP56 DACs already support these speeds to a reach of 3 meters. While longer-distance 50 Gb/s deployments will need AOCs or transceivers with fiber cabling, enterprise data centers using a ToR approach are well positioned to support increasing server speeds with SFP-based DACs for several years to come.

Just as many of today’s enterprise data centers use QSFP+ to 4 SFP+ or QSFP28 to 4 SFP28 breakout DACs or AOCs to support 4X10 or 4 X 25 Gb/s server connections, 50 Gb/s server connections will also be supported using breakout assemblies. For example, a 200 Gigabit QSFP56 to 4 SFP56 DAC or AOC will support 4X50 Gb/s server connections. Time will tell, but as 400 Gigabit starts to enter the enterprise data center environment, we may even see the 8-lane QSFP-DD form factor used to support 8X50 Gb/s or 4X100 Gb/s server connections with up to about 2 meters supported by DACs and longer distances via AOCs.

With a full line of SFP+, SFP28, QSFP+, and QSFP28 DACs and AOCs, as well as multimode and singlemode fiber and all categories of copper structured cabling, Siemon can support current enterprise data center speeds for both switch-to-server and switch-to-switch links-no matter what length you need. And you can rest assure that we’ve got on our eye on the enterprise data center market and will be ready to support your future 50 Gb/s server connections with a full line of SFP56 DACs, AOCs, and breakout assemblies.