​

By that time we had received both the server board and the main board. Pierro was working on the boot sequence and the BIOS for the Intel chips. Jérome was working on the fan and temperature control in the firmware as well as designing the chassis metal work in CAD tools. I was enabling the network features of the network chip and doing some of the firmware code. Greg was already doing the traces for the next generation in his EDA tool. Four people working on different pieces and bits to make the hardware work.
​

Even though it was a new design both in terms of the server chip and the network chip, there were a lot of behind the scene improvements that were done compared to the first generation. Some of the issues we had found on our own and some from the users feedback.
​

The NAND chips were swapped for NOR chips. The fan and temperature control board was coalesced into the main board. The network chip had a dedicated management network instead of being inband. The chassis form factor was switched from telecommunication width to standard rack width. Other small improvements that came from the experience of the first version were also made.
​

Most importantly, I think, the biggest changes were for the end customers. The C1 server had 2 gigabytes of RAM memory and no local storage. This was an issue because it didn't handle a lot of different types of workloads. For the C2 server, we added ways to be able to better customize the servers. We could use three different versions of the Intel chip. Instead of soldered memory, we used standard memory sticks. Lastly, signals were drawn from the server board to a special place on the main board where we would vertically stack the storage drives.

A lot could be reused, but a lot more new things had to be qualified. Every of these hardware features had to be thoroughly implemented and tested. To do that we had to write software in multiple locations. Whether that be firmware code, operating system code, application code or tooling code. Then, we had to devise tests to check that everything was correct. For the SD-card slot, it was simple enough, but to test that the drives linked correctly in version 3 of the SATA protocol and that there were no errors when we pushed a lot of data, was a bit more difficult.
​

Whatever we worked on, most of the time the only information we had were the electrical traces, the multiple dozens of components PDF data sheets, and the underlying protocol information. Sometimes those data sheets were not very verbose, so you had to guess and check.
​

The impact of both the design and the physical changes was humbling. I could see the changes, I could understand the why and the end result, but the process of accomplishing the change, I could not yet reach.

​
We huddled around each time we received a new revision of the bright red boards. Looking at the components soldered and flipping around the boards to check for issues. We closed ticket by ticket for each feature. Slowly moving forward centimeter by centimeter. This was the way.
​

​

If you have missed it, you can read the previous episode here​

​

To pair with :

• Springtime Linn - Clark
• Le Comte de Monte-Cristo (The Count of Monte Cristo) by Alexandre Dumas