Hacker News Reader

Comment by Aromasin | Hacker News Reader

As someone who designs networking chips for a living it's always entertaining to me when people get very excited over their new fiber optic cables, despite the fact their PHY/MAC doesn't support much over 1GbE. You can use Copper all the way up to 10GbE without too many issues. Fiber has a theoretical maximum of about 44Tb/s [1]. It's going to be a long time before we need to worry about advancing the link technology itself.

We're getting a lot of the big telecom companies churning out a lot of 400GbE designs now with custom silicon/FPGAs though. 800GbE and 1.6TbE specs in the works by IEEE too [2]. It's amazing how far we've come since the 1000BASE-T back in 1999.

[1] https://www.nature.com/articles/s41467-020-16265-x [2] https://www.computer.org/publications/tech-news/insider-memb...

View on HN

Not sure I follow, isn't the reason you are excited about fibre that it implies a good connection to the "backbone". Compared to DSL, cable or wireless which are is really limiting even in the 50mbit range (in practice).

And for the distances we talk about you kind of get fibre anyway since it makes the most sense. But many do have a copper connection to their fibre, just the last stetch.

I was just making a commentary on the traps some of our end-customers fall into, when they upgrade the connection to fiber thinking it will improve their network speeds when the protocol is in fact the limiting factor.

What protocol? I can get 10 Gbps internet service right now.

Of course; telecoms are always pushing the leading edge of data transfer would need fiber for 10/25/40/50/100/400GbE. Some of our customers are still making designs using Triple Speed 10/100/1000Mbps though. For embedded use cases that's good enough, but they still often needlessly spend more on fiber cables, regardless of necessity.

Fiber is very cheap. Easily cheaper than good copper cable if you're aiming for 10 Gbps or later.

The hardware less so, but even then very much affordable.

If you're planning for the long term and think that running new cabling everywhere would be a huge pain, or there's some sort of concern with interference, I would definitely go with fiber.

I'd love to agree, but there's a whole list of companies I can name and shame where upgrading was, and is, simply a waste of time and resources for everyone involved in the design. It'll do the same speed regardless, with identical error rates, because the protocol hasn't changed. It's simply marketing guff that they can slap on the same product. The "gold-plated HDMI" scam with a fake mustache.

I think you've just got some specific situation that's not quite obvious from your comments. It's not at all clear what's being upgraded, what kind of equipment is involved, and why you can predict the effort will be useless.

End users traverse the same core equipment no matter what the last mile cable media is. Newer fiber builds may have more efficient core gear or have more capacity but this is all limited by the ratio of users served by a particular backhaul link, and also the topology and configurations made by the operator. Large networks are a challenge for some.

Fiber is a lot less susceptible to water intrusion but still has the same potential for outages due to cable damage along roadsides and on aerial poles with power lines.

Of course, but that doesn't change that fiber is typically a massive improvement to anything else.

Because that means proper last mile connectivity where DSL suffers and cable has poor latencies. There is also less contention in the last mile.

It's a bit funny, but it's really a question of distance more than medium.

For in-rack, DAC/Twinax is just copper and good for 400G, and cheaper than separate optics and fiber.

And, DAC has lower latency and uses less power than 10GBASE-T even at the lower 10Gb speed (e.g., due to latency you don't want storage over 10GBASE-T). And, DACs contain fewer parts that can fail than optical transceivers (esp. short passive DAC cables) so more reliable (as long as not physically manipulated) than fiber + optical transceivers.

DAC cables are a bit harder to route than fiber though.

That should read:

And, DAC has lower latency and uses less power than 10GBASE-T (e.g., even at the lower 10Gb speed due to latency you don't want storage over 10GBASE-T).

(Not that anyone is still reading this old article or would likely care if they were)

As someone who has rigged up his house with MOCA everywhere, I just want to say that coaxial cable is an impressive piece of ancient tech. It's just a piece of copper that vibrates at specific frequencies but the sheer level of multiplexing that goes into a single wire is impressive, simple, and reliable.

Crossing clock domains is a very interesting topic. For these ultra high speed connections, does it simply boil down to a really big fifo? One could easily imagine a tree of fifos that are interleaved by some scheme.

A large DCFIFO array/synchronizer chain is part of the solution yes, but it's not quite a one-size-fits-all solution. Depends on if your signals are synchronous or asynch, and if it's a pulse or not in a single bit solution. For multibit it depends whether it's static data, whether a transfer is required every cycle, if the data needs buffering, if the data is a counter, and whether all the bits must arrive in the same cycle.

Reply on news.ycombinator.com