The next generation of 25/50/100GbE Ethernet hasn’t been finished by the IEEE but no one seems to care much as products based on specifications of the 2550100 Alliance have been built into silicon and vendor brand are shipping switches today. While this highlights the deficiencies of the IEEE it also highlights the demand for faster Ethernet speeds.
The Hardware is Here
Vendors have announced a number of products that support 25GbE.
- Dell has the Z9100 – Dell Networking Z9100-ON Spec Sheet
- Mellanox has a new product family called Spectrum based on their own next-generation of silicon that isn’t just 25GbE but also 64x50GBE units shipping today.
- Arista has announced a new family of Arista 7060X, 7260X and 7320X fixed-leaf and modular-spline form.
Behind this move is the release of the Broadcom Tomahawk chipset which offers 128 x 25GbE channels that can physically presented as 32 x 100GbE, 64 x 40 or 50GbE or 128 10/25GbE ports depending on the breakout cables and software licenses purchased from vendors. Other vendors should be close to announcing products in the near future – here is Cisco’s statement
The cost of 25 GbE on a per port basis is about 1.4 times that of 10GbE. Some vendors may charge more but the cost of manufacture of 25 Gigabit hardware is about the same price at 10 Gigabit. Vendors will obviously be hoping to charge a bit more and generate some extra profits but my perception is that whitebox/whitebrand is setting the floor price for switch hardware at a much lower price point than we have seen in the past.
Skipping 10GbE ?
In 2010, I said the following about 10GbE:
I’m expecting 10GbE to get slow, progressive adoption over the next three years. It’s not an industry revolution, and not enough people need to increase bandwidth to drive rapid. The only use for 10GbE in the Enterprise is for Data Centres where Storage and Virtualization/Blade Servers are driving adoption.
10 Gigabit adoption has been slower that anyone predicted. Customers have been slow to replace networking hardware. Whether its because networking is risky and hard to upgrade, or because prices are too high is unclear. But it might be worthwhile skipping 10 Gigabit entirly:
- Customers are continuing with 1 Gigabit. Vendors are somewhat concerned that sales of 10Gigabit have been slow. Customers often highlight that 10GbE is too expensive and not worth the upgrade.
- Todays servers can oversubscribe a couple of 1Gigabit Ethernet links but only just. While virtualization and IP Storage do create bigger network loads not everyone has those problems.
- Vendors like to talk about investment protection during the sales process, and 25GbE can be attractive there. In most cases, investment protection is an illusion but the
enterprise IT humancan’t resist the allure of safety.
- The products above support 50GbE and 100GbE in addition to 25GbE. This will be attractive to customers who want to believe that expensive products should last for 10 years to get a return instead of buying low cost products that can be replaced quickly.
25GbE & The Campus
Although the first sales of 25GbE will be in the data center (big public clouds, HPC Ethernet) I can see that Campus Networks will be keenly interested in 2.5 times the bandwidth for 1.4 times the money. Add in the flexibility of having the same physical ports and cabling to be used for 25 or 50 or 100G speeds on the same device and there should be demand in the backbone. As WiFi speeds increase and deployed density means more devices the demand for backbone bandwidth seems good.
On the other hand, the campus seems locked down for now. There is limited appetite to for a large scale transition in the closet.
Factors Against 25 Gigabit Ethernet
- Servers. Or more specifically, network adapters. Most x86 servers now ship with 10GbE on the motherboard and using 25GbE will require additional NICs to get 25GbE.
- Getting people to recognise that 40GbE to the server won’t happen because of the cost.
- Breakout cables are required. It took several years for the use of SFP modules be understood by customers. I expect that customers will be equally slow to buy breakout cables until awareness, experience etc have
proventhe product. (Yes, its ridiculous but it will be an issue).
The EtherealMind View
On one hand, 10G Ethernet is finally becoming the normal. After a decade of marketing efforts by vendors the transition is showing signs getting serious and the single largest factor is that servers are coming standard with 10GbE NICs.
That said, most companies only need a couple of hundred 10GbE ports in the data center. So maybe a move to bypass 10Gigabit has some chance.