Think about it, they travel straight through earth at a speed which apparently is even faster than photons :-)

Seven, we are in agreement.  My points were purely selfish (I build EAD's).   End-to-End is always considered.  However in todays world you only controll what you can for your network.  There are plenty of off-net providers whom have no choice as the network is from another provider. 

Having been in the telecommuntications business I know copper paths were longer then 100's of feet, but do do so required special tuning and was frequency limited.  Just stating the obvious.

Fiber optic index of fraction is the reason why light speed varies in optical fiber.  The bigger the index the slower light travels.  Theorically the speed of light is 300,000 kilometers per second however most SMF work around 200,000m/S. 

Delay is always present regards of medium, in the case of this article low-latency is used as a value add for profit.  Low latency paths require special equipment and the shortest route, which some customers are will to pay premiums for.  Generally financial institutions are the benifactors.  There is some move towards using low latency equipment in cellular backhauls as well as wireless is saddle with higher latency (RF) and are now requiring more real time services.  

One important factor left out of this is the fact that CoS and QoS are generally left out as processing doesn't allow enough time for packets inspections.   This generally means the provider has to make sure they configure the circuit appropriately and not have any other services restict or affect this circuit.

Rush,

But if you are a carrier building a wireless network you actually have to construct a full end to end delay model using every transmission line and every piece of equipment.  The problem with copper is that it is very nice for a few hundred feet of low latency, but not much more than that.

The other thing that is almost always left off is physical propogation delay.  Speed of Light in a vacuum is about 1nsec/foot (which is about 5280 nsec/mile or about 5 usec/mile).  So, in an end to end model there is significant latency in just going from point A to point B.

seven

brookseven - I was only referring to EAD's, glass and copper not other devices.  Yes you are correct when using active devices there is considerable delay FEC being one of them. 

I did make one mistake in that I should have said 'can't' instead of 'can'.  Most switches, routers and aggregators are in the order of mS not uS. 

Rush,

You have missed the potential modulation delays that are built into DSL modems. Interleavers, for example, can take 10s of msec.  So, yes the transmission delay is less IF and ONLY IF all the FEC and other systems are turned off.

Those Ethernet numbers for switches are independent of the type of medium that they are transited over.  Those are for the switches.

seven

Low latency for Ethernet Access Devices are measured in the factor of uS instead of mS for various frame packet sizes.  Most people don't understand that glass has more latency then copper as a medium; thats not saying we should revert to copper only physics.  

Generally sub 3-4uS for all frame sizes (inlcuding jumbo) is considered state of the art for Ethernet devices as most routers, switches or aggregation devices can do this - generally a cut-though mode or very fast FPGA is the reason for this.

The Dragonwave guy (about 1 minute into the video) seems to have a disconnect with what others mean by "low latency networks" -- where I'm from, latencies in "low latency" networks are measured in *nano*seconds!

Just to put that in perspective: light travels about one foot per nanosecond in a vacuum -- less than that in a medium. I could be wrong, but I doubt that a hundred nanoseconds more or less latency are relevant in a "mobile backbone" or "LTE deployment".

Low-latency networks matter most applications such as High-Frequency Trading (HFT) and supercomputing.