The network chips we used had many more features that we had implemented in software. As the complexity of how the devices were used and set up in production, it raised the rationale to use more and more of the advanced chip features.

It’s one thing to have routing capability, it’s a better one to have multipath routing capabilities. The multipath capability enables sending packets, data split up in small portions, through multiple links. On the compute hardware, we had four 10 gigabit links, it would be better to be able to use multiple links on the same path. That would bring the capacity to 20 or 40 gigabit if we aggregated.

Unlike bonding capabilities, which I had also looked in, it had a much better compatibility with various devices. The way this worked was that you configured the chip to send a portion of the traffic through a path depending on parts of the data of the packets. It would create a hash based on specific parts of the data to be able to deterministically send the data to the correct link. This required some rework in our code but the hardest part was actually testing it out.

That required producing the setup and manually checking which link had the packet exited from. Many small details had to be accommodated. The netlink interface had to be patched, the routing code had to be patched, the interface had to be patched… And everything tested to be accurate. This simple user feature was complex behind the scenes.

The next one was tunneling. This was also a feature available in the chip. I started looking into it and preparing the implementation but the doubts from the SerDes eye encounter got stronger and stronger. I had trouble understanding how it worked and unlike every previous encounter with things I initially did not understand where I would push through, this time I did not.

Hindsight is a very powerful tool. Unfortunately it only works on the past. It is very clear to me that multiple things contributed to my state at the time. The lack of self confidence had caused the SerDes eye failure which, in turn, triggered this whole “I do not belong here”. I thought this was the reason why I had not been assigned EDA work yet. This was the reason I could not do the tunneling or understand the very dense LACP specification documentation.

Or at least, I believed that. It wasn’t rational. It wasn’t based on the track record. It wasn’t based on the actual interactions and work friendships I had. It was a belief that I was an imposter. But it wasn’t true.

In that mix, there was also four years of metaphorical blood and sweat. Four years of pushing forward, pushing again and again my knowledge and skill boundaries. A never ending learning curve that kept rising steeply. I was tired. My brain was saturated. I hadn’t taken a proper vacation or break in these four years. That had a cost. But I could not understand it, not yet at least.

Changes were about to happen.

​

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

​

To pair with :

• Bad girls - M.I.A.
• Big Planet by Jack Vance