Will the 5G core network technology offer lower latencies?

[simeonz]

Reading about 5g and its many promises, I ponder on the sub-ms latency aspect. This is the latency to the nearest cell tower, obviously. But if they change the core infrastructure as well (which they seem to be planning to do) and possibly make it wireless, it may allow avoiding some of the routing congestion, simply because the increased connectedness between the nodes will enable improvements in the distribution of routing workload.

Does anyone have any idea if the delays of the core traffic will drop significantly with 5G once it gets widely deployed (to the limiting point of 3.3 microsec per kilometer radio speed of course)? This is a bit off-topic, I know, but it may have impact on the architecture choices for OSes (and the significance of the cloud as an alternative).

[LtG]

I'm not sure what changes in the "core infra" you are referring to and how they affect routing congestion?

What kind of delays are you referring to? Ballpark figure?

Also, I'm pretty sure they already use wireless links from cell towers when it makes more sense, but in urban areas I would expect using wires will continue. Though haven't really studied that.

And remember, electricity doesn't travel 1km/3.3usec, it depends on the wire how fast it travels. It can be significantly lower, like less than half of speed of light in a vacuum (~300 Mm/s). Not sure if really relevant here, but though I'd mention it anyway. Of course other devices in the path (switches, routers, amplifiers, etc) tend to add quite a bit of delay also.

https://en.wikipedia.org/wiki/Velocity_ ... ty_factors

[simeonz]

I'm not sure what changes in the "core infra" you are referring to and how they affect routing congestion?

What kind of delays are you referring to? Ballpark figure?

A ping from Europe to Australia (au.archive.ubuntu.com) takes 300-400ms. The geodesic distance from Europe to Australia is somewhere from 13Mm to 16Mm. The speed of light in optical fiber is about 200Mm per second. So the roundtrip time for the signal itself should be about 160 ms on a straight line or about 220ms on a "diagonal" (non-optimal path, steering up to 45 degrees from the geodesic). So the remaining 100-200ms delay, I assume, should be routing latency. Which seems strange to me, considering that top of the line core routers are capable of sub microsecond routing table lookups, but this is only when the packet doesn't have to wait in a queue. That is why I assume that the remaining 100-200ms are effects from congestion or deep packet queueing at the core internet routers. It should also be mentioned that the icmp packets are handled differently from conventional payloads, so the ping latency may be misleading me. I have to test the latency with some actual tcp or udp application protocol.

Also, I'm pretty sure they already use wireless links from cell towers when it makes more sense, but in urban areas I would expect using wires will continue. Though haven't really studied that.

I believe that what you are referring to is the backhaul traffic, which connects the access sites to the internet core routers. But the core infrastructure, I believe, is the same one that is used for wired subscriptions. I speculated in my previous post that in order to take advantage of the reduced access latency of 5g at the access sites, the network operators would have to somehow upgrade the core infrastructure as well, or otherwise this point would be moot. If they did deploy wireless infrastructure for the entire end to end communication, this would allow them to form the intermediate connections dynamically and may be reduce the congestion and queueing effects. Because, when it comes to wired communication, the variety of electrical paths available to be taken by the payload on its way to the destination are inherently restricted. This is largely speculation on my part though. If someone here has already read something about the transition and could pass a few words of wisdom on how the overall latency will be handled by 5g, I would appreciate the hint.

And remember, electricity doesn't travel 1km/3.3usec, it depends on the wire how fast it travels. It can be significantly lower, like less than half of speed of light in a vacuum (~300 Mm/s). Not sure if really relevant here, but though I'd mention it anyway.

I said 1km/3.3usec, because for wireless, the signal will be traveling through air. You are right that for optical fiber, this is about 2/3rds the speed.

Of course other devices in the path (switches, routers, amplifiers, etc) tend to add quite a bit of delay also.

Exactly the kind of latency I was wondering about.

[simeonz]

To add a few remarks to my previous post. My estimate of the routing delays was admittedly very back of the envelope and in retrospect not very useful. I would have liked to take into account the length of the actual submarine fibers from Europe to Australia and compare it to the ping times, but I couldn't compute the lengths for relevant segments of the SEA-ME-WE 3 connection. The total length is known, but the Europe-Australia segments are less then that.

On the other hand, some intercontinental links with known pings are essentially direct, so I can infer the routing and device latency from that.

Interestingly, the Perth to Sydney gives a perfect ping time of 46-47ms. Considering that the submarine connection (INDIGO-Central) between the cities is 4,850 km long and that the speed of light in optical fiber is slightly above 200Mm/s, this result is very tight (considering the signal will have to travel both ways).

On the other hand, the Bristol to New York ping time is 84ms and is famous for being excessive. The submarine connection (Tata TGN-Atlantic) should be about 13,000/2=7,500 km long in one direction, which means that the latency is about 10ms greater then optimal.

Still, I confess, I overestimated the effect of device delays. It appears to be up to 10% of the total latency in general. Whether this is the case for local serving as well (such as by CDNs), I don't know. You can get ms latency for intracity pings and the signal travel time is much less then that. But this may be skewed by many factors, such as the speed of personal equipment and what not. Essentially, I have to concede that the premise of my question was probably wrong for global access. Routing devices do play a role, but the signal latency trumps processing delays. For local access, I am not sure.

[LtG]

Are you considering the 5G tech for long distance links or short distance (as in cells)? For short distance to be viable to route your traffic from EU to Australia would mean you'd probably have to have billions of towers on the planet --> not viable. I'm not sure if the 5G (haven't studied that much) contains stuff about long distance links.

If you're worried about ICMP being mistreated, then you could also try TCP SYN-SYN/ACK timing to see what the "real" latency is. You could just try to telnet (or netcat/nc) to some IP address and port and tcpdump/wireshark the traffic and check the timestamps.

I believe the low latency applications of 5G are intended for onsite. The idea being that factories can take more advantage of IoT with more or less guaranteed low latency, this would reduce costs and maybe make factories more flexible. Cabling if often somewhat difficult to change after the fact, if it becomes a non-issue then it's possible there will be more innovation. Once you go offsite, no matter wired vs wireless, you start to pay the c (speed of light) latency penalty, and immediately you're out of the <1ms game. So as I see it, the 5G latency is there to allow sites to go entirely wireless.

If they did deploy wireless infrastructure for the entire end to end communication, this would allow them to form the intermediate connections dynamically and may be reduce the congestion and queueing effects. Because, when it comes to wired communication, the variety of electrical paths available to be taken by the payload on its way to the destination are inherently restricted.

The airspace is a shared medium, so it too has its downside. Plus, you still need to build the link towers, so I'm not sure if there's much gain here. If anyone knows of any plans, I too would be interested to hear.

[AshleyEchols]

What kind of delay do you mean? On 4G networks, the delay range is somewhere between 50 and 100 milliseconds. 5G networks will reduce this delay to 1–5 milliseconds.

[simeonz]

What kind of delay do you mean? On 4G networks, the delay range is somewhere between 50 and 100 milliseconds. 5G networks will reduce this delay to 1–5 milliseconds.

Depends with what you communicate. For devices very close - such as IoT (as LtG noted), the delay can be small. But for peers on a different continent, the speed of electromagnetic signals is a hard limit on how fast you can reach the other end.

[simeonz]

Are you considering the 5G tech for long distance links or short distance (as in cells)? For short distance to be viable to route your traffic from EU to Australia would mean you'd probably have to have billions of towers on the planet --> not viable. I'm not sure if the 5G (haven't studied that much) contains stuff about long distance links.

I would say, I agree overall. For medium distances, they could try broadcasts from high altitudes, similar to aerial TV, but for intercontinental transmissions, they would have to bounce off satellites. This is even worse latency-wise (not to mention - does not avoid routing), so the wireless route is not viable overall. And as you also mentioned, wired offers easier scaling.
I think you were right - intercontinental communication will have to remain fibre I think.

I believe the low latency applications of 5G are intended for onsite. The idea being that factories can take more advantage of IoT with more or less guaranteed low latency, this would reduce costs and maybe make factories more flexible. Cabling if often somewhat difficult to change after the fact, if it becomes a non-issue then it's possible there will be more innovation. Once you go offsite, no matter wired vs wireless, you start to pay the c (speed of light) latency penalty, and immediately you're out of the <1ms game. So as I see it, the 5G latency is there to allow sites to go entirely wireless.

For IoT in the public space and mobile IoT, I can see the point. But for privately owned on-site networked devices, wouldn't 802.11 wireless be better then GPRS. I mean, why would you pay the network operator for using equipment that you can own yourself. Although, sub-commissioning infrastructure may optimize staff and maintenance expenses.

There is also something called edge computing that is purported with the latency shaving of 5g. Geo-distributed cloud of sorts. For underdog developers like the members of this forum, this is not such great news.