This week has been a whirlwind, I am writing this newsletter a mere 4 hours before it’s scheduled to go out, which isn’t doing favours for my anxiety. Normally I jot down ideas as I go during the week.

Next issue is going to have a longer article about some of the AI sentiments and my thoughts on them, so stay tuned for that!

This week’s theme was around snowflake IDs, time orchestration, fiber optics, quartz movements, and rack sizes in industrial and home.

Here’s what I learned this week:

Aggregation switches are cool

An aggregation switch takes lots of physical network links and aggregates the traffic on all of them into a single link to send upstream.

Conversely it takes the data on the one link, and routes them into the individual smaller links.

The one link should be significantly faster than the others, like 10G vs 1G.

Data on fiber optic networks is weird

On your normal copper wire, you have digital signals encoded as voltage changes between 0V and 5V.

In a fiber optic cable, it’s light. There’s Wavelength-Division Multiplexing that does the majority of work. This means that per unit of time, a lot more data can be transmitted because you have more links (one per colour).

Aggregation on fiber optic cables is just a prism

Fiber optic aggregation switches have a prism that splits the incoming light into 64, or 128 outgoing light links. That also means that everything that comes out of the one link gets sent to the other 128 as-is, as it’s just a prism. The light doesn’t change.

Addressing in light is encrypted

The aggregation means that your fiber-to-the-premises internet gets ALL the light for your house, and everyone else in the neighbourhood.

Your ONT (the device you got from your internet service provider - Optical Network Terminal) takes all of that light, filters the data that’s intended for it, and ignores the rest.

Intended for it means the data is tagged with an ID corresponding to that terminal device, and encrypted with a key that’s on that device.

That means you can not replace the device yourself, the ISP needs to know which device needs to receive the traffic so they can address and encrypt your traffic.

Sending data back from many fiber optic lines to a single trunk line needs to know the exact distance between your ONT and the aggregator switch

Because it needs to know exactly how much time it will take for the signal to get to the switch in nanoseconds. It knows the speed of light.

Data going back from 64/128 fiber lines to a single one will repeat that data across all of them, so the only way for the switch to not get garbled overlapping light signals is to tell each ONT when they’re allowed to transmit.

Enter Time Divison Multiple Access. Basically each client gets a few ms to transmit their data. That data needs to travel down an optical line. That optical line will be variable length, so in order for all the signals to arrive at the exact ms they are allowed to transmit, each ONT needs to know how long their line is. Which they measure.

Slightly different, keystone modules are a standard

It’s the size and fixing of different things, which makes paneling easy: in racks you can have keystone jacks, and they you can put however many keystone modules in them in whatever order you wish.

Your house’s phone line, ethernet plug, or antenna in the wall are most probabl keystone modules too.

Racks come in two sizes in width

10 inch for smaller desktop lab devices, and 19 inch for your industrial or enterprise devices.

There’s also an older 21 inch for telecom devices, but that’s legacy.

Fun fact, the 19 inch is an actual standard: IEC 60297 Mechanical structures for electronic equipment – Dimensions of mechanical structures of the 482.6 mm (19 in) series. You can buy it for only 210 swiss franks, see my previous newsletter on why these cost money.

The 10 inch one is more of an agreement that a bunch of people on the internet said “yeah we’ll adhere to it,” but there’s no standard behind it.

There’s a good reason why hi-beat quartz watches are rare

There’s only one movement that I could find, Bulova’s Precisionist movement. Instead of the normal 32768Hz quartz crystal, this one uses 262kHz three-prong tuning fork crystal and a stepper motor that drives the second hand 16 times a second.

Mechanical watches have this by default, because the escapement oscillates fast on its hairspring anyways.

Reason quartz watches don’t move their second hands faster than once a second is energy. Their batteries and capacitors are just not strong enough to power motors that tick that often.

A cheat to get more out of a snowflake ID timestamp. Or any timestamp

Snowflake ID is just a single 64bit number composed of three parts:

• 41 bits for time

• 10 bits for machine ID

• 12 bits for sequence within the same µs on the same machine

That 41 bits at the front is just barely enough to encode current time in milliseconds. As approximation, May 2026 is roughly 1,748,000,000,000 in decimal, which does need 41 bits to encode, and its max value is 2,199,023,255,552, which is April 2039.

You can get an extra 69 years! (nice!)

Epoch works by offsetting milliseconds since January 1st 1970. Absolutely nothing says you need to keep THAT date as your 0. If you’re launching a service, you can treat 1st June 2026 as your 0, in which case:

• 1st June 2027 – 31,536,000,000, and

• max date is 2,199,023,255,552 ms since 1st June 2026 – March 2095

All right, this are all the things I learned via AI. It’s been a lot. I have no idea what I’ll be doing next week around it. 😅 It will be weird and niche, as always. Until then!

Loveyoubye! ❤️