1st generation network equipment

​
Building computer hardware is an interesting business. The entry price is horrendous. It takes months of careful planning before you can get any physical​devices out. Of course, the physical device is not enough. You also need all kinds of software for it to work. The First generation required two firmwares, two Linux and a set of daemons involved to get to a user land shell on a server. The second generation became three firmwares, one bios, two Linux, and a bigger set of daemons to get to a user land shell on a server. How you manage the sequencing dance determines how fast you go to market.

The mistake a lot of people make when thinking about a device, is thinking the product is the hardware, it's not. It's a mix of hardware and software. I could give you all the physical hardware, and even if I gave you schematics, you would not be able to use it without writing a lot of software. Think of it as an iPhone which is not hardware but a product. You are buying into an ecosystem where there's an operating system, base applications, an app store and a lot of existing apps created by third parties. The physical device is almost useless without the software that goes with it. The software part takes people and time to write, and they receive compensation in exchange for their work. In essence, you buy the sum of the time spent on designing and testing the hardware, the sum of the cost of the components, plus the sum of time spent on writing the software for all the ecosystem.

Although once you pass the entry price, you start to have more and more opportunities. Think of it like an investment. It's a lot of CapEx, then it compounds into more knowledge, more experience and more opportunities. One of these was a networking device. Existing brands such as Cisco or Juniper equipment are expensive. You cannot build a cloud service or server hosting company without networking. The entry price to make networking equipment is extremely high. To make it pay out, you need either to sell it to volume or have sufficient in house volume of network equipment that the price of the device becomes attractive.

But, what if you already have 80% of the R&D done and you buy a sufficiently high volume of this equipment? Does it become interesting to have your own hardware? Scaleway was exactly in this situation, by just redesigning part of the second generation and adding a few components, we built a router. It had 48 ethernet ports (1 gb/s) and four SFP ports (10 gb/s). It did not require a lot of software changes as it was the same family of ASIC, so it was just a few configuration changes.

Still, we had to do some qualifications. For each port, we had to test signal integrity, signal correction and signal strength as well as being able to manage the chip that handled the ethernet signal. Nothing too complicated at that point; however, the complicated part was exactly what I mentioned before, the product. Even though we had the hardware and it worked, we had a bunch of tools to control it, what we did not have was the ecosystem. People did not buy Cisco and Juniper for the hardware only, they bought it for the ecosystem.

How could we lower the friction to adoption? Well, the answer was asking the co-workers about what was important in their day to day operations. Looking at the tools they used and making sure that we supported them. Dynamic routing via open source tools is one of the examples of where implementing netlink made them save a lot of time as they could use the same workflow as their existing systems.

One thing that came up was open computing. I think they had started to work on a common cli or interface for network equipment. It was very early work, and we decided it was too much work for us to start adapting our stack to fit into their model. In the end, it's a balancing act. How much effort do you put for a standard and how much results you'll get from the effort you put into it.

​

To pair with :

• Golden Skies - Mono/Poly
• Redemption Ark by Alastair Reynolds