-1

Let's say I have a 1Gbps bandwidth NIC. Can I mathematically derive what would my network latency be just from the bandwidth number?

Always_Beginner
  • 99
  • 1

2 Answers2

4

These are 2 different concepts.

More bandwidth means carrying more stuff in the same time (provided the whole link from your NIC to the server NIC is fast enough).

Network latency is the time it takes for data to go from your NIC to the server (destination) NIC (usually linked to the distance between the two, but not only).

Déjà vu
  • 5,408
  • 9
  • 32
  • 52
0

Bandwidth and latency are not tied together, so there isn't a mathematical way to compute it.

Bandwidth is how much data you can cram onto the wire. Latency is how long takes the data to reach it's destination.

Think of it like this. If you have a 100mbit connection between you and the destination, but that destination is only 10 feet away, the data would get there very fast. But, lets say the destination was 100 miles away. It would take longer to get there. Add on top of that delays added by network equipment between you and the destination. Even 100mbit routers and switches add a tiny bit of delay in order to process the packets that flow through them. So, if you bounced through two switches and two routers, each adding 1ms of delay, you'd have 4ms delay even at 100mbit.

mikem
  • 408
  • 2
  • 6
  • Your answer is only partially correct. Bandwidth and latency are tied together, particularly minimum latency - and inferences can even be made about relative latency. **The issue is there are confounding factors like the speed of light,interleaving, buffers, digital, switching times**. If you replace a 10 megabit connection with a gigabit one (all else being equal) your latencies will drop - even your minimum ones. I've correctly predicted an up streams fibre connection being upgraded - more then once - based on the latency minimum. Congest a circuit and you can make additional inferences. – davidgo Aug 31 '20 at 07:03
  • Seems unlikely to me given the sheer number of variables that can impact latency -- cable quality, signal interference, equipment-imposed delays, distance, etc. But I've been wrong before. Can you share the formula to answer the OP's question? – mikem Aug 31 '20 at 07:19
  • Thats the point. There is no single formula, but there are correlations, so to say "Bandwidth and latency are not tied together" is not correct. What did it for me wasn't a formula - it was simply using Nagios to graph latency (Mainly between 2 cities about 600kms apart. The reality is that, save for utilisation, the variables remain fairly static. Doesn't hurt that I've been in the industry since 64kbit circuits fed an entire ISP, so I've a wealth of gands-on knowledge to draw on – davidgo Aug 31 '20 at 07:55
  • I'll have to respectfully disagree. Correlation does not equal causation, and without specific evidence or formula, I just don't see the relationship. You can increase or decrease interface speed (ie 10mbit/100mbit) on a crossover cable and not see any change in latency. Same distance, same cable, same equipment, only a negotiated speed change. Do a similar test, but significantly change cable length for the 100mbit test and you will see higher latency on the faster link. Of course, these just are our opinions. Perhaps there are others out there that would like to comment. – mikem Aug 31 '20 at 08:11
  • Ill just refer - https://serverfault.com/questions/276651/network-latency-100mbit-vs-1gbit#276652. Certainly I acknowledge cable length can impact performance. – davidgo Aug 31 '20 at 08:27
  • Interesting read. Unfortunately, like our conversation here, there isn't anything conclusive. Lot's of what-if's, opinions. and some testing with switches in the mix and varying payload sizes... which could affect NIC and switch performance. Still, very interesting. Thank you. – mikem Aug 31 '20 at 08:59