Discussion in 'Networking, Telephony & Internet' started by Smokin Whale, Sep 21, 2014.
Haven't considered those, to be honest wasn't aware of those! Will check them out!
technically, why does copper have increased latency over fibre (SFP)? is it the speed of electrons over copper vs photon in glass? or something in the tranceivers?
at a guess, DAC is an electrical signal, whereas fibre is light.
IIRC light moves 3x faster than electron movement
It doesn't in token ring configuration
its because of the agreed encoding standard funny enough. nothing to do with copper vs fibre transmission speed.
nearly 10 times worse. :/
i wonder why fibre can get away with simpler encoding? less error rate? or just legacy compatibility reasons? that would be the next google...
This article is rather old but perhaps still relevant... https://www.datacenterknowledge.com...-benefits-of-deploying-sfp-fiber-vs-10gbase-t
At huge distances, what you did assert is correct, the electric wave through a good conductor travels at near speed of light, BUT as the current laws (well accepted theory) of physics states, nothing travels at the speed of light except for mass-less quanta/photons.
"If an object tries to travel 186,000 miles per second, its mass becomes infinite, and so does the energy required to move it. For this reason, no normal object can travel as fast or faster than the speed of light." ...ref.
The signals going through the copper cable though are electric waves and not electrons moving through the medium at near light speed. An example would be to hang 10 ball-bearings on a horizontal stick with equal length strings. Pull one of the balls on an end horizontally and let go (let's say left side). The ball on the other end then moves (right end side) as the energy wave moves through the balls to the last ball on the other end. In effect hardly any balls move, but the energy wave moves through all the 8 balls in between at super speed!
Even better on a good well insulated electric conductor. The electron clouds in the metal crystal lattice conduct the energy waves at near light speeds for good conductors.
Sorry didn't realise you had replied to the question quoting the exact same link! Lol. Easiest result from Google I suppose.
I'm curious as to what possible usecase the lower latency would make a difference in
We are talking far less than 0.001ms in both situations so...
its one of the reasons SANs and NASs are shitty for small packet access, forcing developers to use latency hiding techniques.
Multiple hops in complex infrastructure to get from point A to B. All the latencies add up.
An example as given in the links above is the routing mechanism of the Tor network
EMR affects electrons in copper, but not light in glass. Outside of physical damage to the fibre, there's not much to get in the way of those little photons whizzing down the glass.
With that assumption in mind, why bother dealing with a bunch of encode/decode and error checking if you don't need to?
I think it's a case where if you're in the top 1% of users, you'd notice it. If you had some insane network-limited / massively clustered storage setup or were ISP scale networking, then sure.
For the rest of us plebs, I don't think it's an issue.
FWIW we use fibre everywhere for 10GbE mostly because it runs cooler and is cheaper with everything factored in. We've sourced fibre cables and SFPs direct from China in years gone by, and more recently local resellers who now do the importing themselves. When doing 10GbE to the workstation, fibre ended up coming out cheaper almost every time for a new rollout.
propagation delay of copper is about 0.95c
actual electron movement is ~m/hour. but that's not what we care about. The electron you push into a cable is not the one you get out, nor does it need to be.
propagation delay of fibre is ~0.66c. so technically slower (and significantly so), but not something you can measure with a ping around your home or office networks. With accurate enough time keeping you can measure the difference, it's in the order of a few nanoseconds over a 100m cable.
But there's much more latency introduced by the PC and any routers/switches in the path (OS, IP stack, PCI bus, NIC, fibre or copper interface, serialisation delays), which add up to microseconds (1000x greater), the cable you use makes no difference in practice.
typical use cases for low latency networking are supercomputing, automated trading systems (stock exchange), and they'll typically not use ethernet as there's lower latency options (infiniband for one, there's proprietary options especially in the supercomputing field), the cable used even in those circumstances is of next to no impact.
ideal location, has yet to materialise (downstairs garage) a ceiling height wall mounted comms cabinet. The temp gets very warm in there and due to the double brick it stays very warm for a long period.
The only ventilation opportunities will be through a load bearing double brick 2 story wall so my best bet is to scavenge ex production industrial POE networking gear (for cameras), 10g copper is a wet dream that would require an old fridge compressor and some propane in that space.
choosing between silicon that caps out at 45c and 75c is simply a no brainer, even if I have to double up on POE switches.
nor was my example 10g, it was both copper and heat related though, which aligned with the threads general direction.
*looks at thread topic*
cmon guys this is home networking kit we are talking about
A lot of Tor network bridges are in people garages all around the world. The Darknet lives in peoples (home networked) garages. Well that's the allegory...
True but your isp latency variation will be a factor of 10000x more than the type of media you decide to use on your 10GBE
True, for us homeboys there's almost no material difference between SFP+ and RJ45 (perhaps temperature would be an edge case as has been stated previously). That's why I bought two Asus XG-C100C (RJ45s) for my rigs.
10GbE copper has to allow for errors. To transmit at that rate over copper is difficult with lots of external factors generating noise that screw up the 10GbE signal. As such the encoding method mentioned above is all about ensuring data that is sent, arrives ok. It has a lot of in built error correction and the like. The extra overhead adds latency when compares to SFP+.
Ubiquiti make cheap SFP+ modules now, what use to cost a fortune is now only $70 for two modules. Fibre can run longer distances, doesn't use as much power, has lower latency and generally works better (is my experience)
I posted several years ago about alien crosstalk in 10g copper and the inability to groom them into the rack / cable tray, although the biggest issue with 10g copper is how utterly unwieldy CAT6A is, I've had CAT6 cables put enough pressure on a switch port that the network drops out if flexed a little, we ended up replacing all of them with cat5e and never had a problem with rack doors ever again.
I've got my rack in my garage and use normal consumer gera. Never had an issue, even when its over 50 degrees in there.