13

This is a general question not related to specific hardware. If we have the following configuration of daisy chained switches:

internet ---- core network ----- switch 1 ----- switch 2 ----- switch 3 ----- switch 4

Do you think the users connected to switch 4 are going to have significant increase of latency compared to users connected to switch 1. In both cases the users access an application in the Internet.

Do you have any example for real case?

|improve this question
  • The questions was for remotely stacked switches which will have one management address and logically will work as one switch. Is there any difference between stacking in chain topology and daisy chaining? – vladiz May 10 '14 at 19:01
  • With stacking you will use the backplane to forward packets in the stack so you have both more bandwidth and less latency compared to daisy chaining. In your case I would not expect it to be a problem but as I mentioned in other comment it can still make sense to connect switches to core depending on requirements for availability. – Daniel Dib May 10 '14 at 20:32
  • 1
    Just pointing out had those switches not been a stack, that's usually not a recommended design. A failure in any upstream switch toward the Internet takes everything downstream out. Uplinking the switches separately to the core would be much better. Now you just have to worry about your single core taking everything out. ;-) – generalnetworkerror May 10 '14 at 20:57
  • @generalnetworkerror For better availability now I find that some switches can be stacked in ring topology. In the example above switch 1 to be also connected to switch 4 . And two of the switches to have uplinks to the core network. In this case the only benefit I see is simplified management (compared to the case without stacking in which every switch has own uplink to the core). Is that right? – vladiz May 11 '14 at 9:55
  • @vladiz, yes, that is how a switch stack would normally be connected; my comment was simply to clarify for other new net engineers that the ASCII design shown isn't ideal unless it's a stack, and in which case you would show switch 4 also connected to your core network. The ring topology in a stack will have two cables into each switch member going out to different switches in the stack. – generalnetworkerror May 13 '14 at 0:00
13

No, they won't for typical internet connectivity. You won't even be able to measure it using typical tools.

For DC environments, we're fighting for every microsecond, but for typical ethernet switches, you get from 5-7 microseconds to 30-50 microseconds per hop (switch). Even 30-50 microseconds is too low to notice; typically latency is problematic for applications or humans as it reaches the tens/hundreds of milliseconds.

Take a look at this Cisco whitepaper

|improve this answer
7

The question mentions that this is not hardware specific, however different platforms have different switching latencies and switching modes.

Cut through switching is the fastest but lets through fragments. Fragment free is the second fastest and makes sure that frames are at least 64 bytes meaning they are not runts which is one of the signs of collisions. Store and forward is the slowest but does not send frames onward that contain errors such as CRC errors.

Your questions contains nothing on distances. If there is a significant distance between switch 1 and switch 4that would have to be accounted for. If they are right next to each other, each switch should not add much latency since they generally forward frames immediately without them getting stuck in large buffers.

Latency is subjective, for a high speed trading firm, 2ms could make a huge difference. For most of us 2ms is not enough to worry about.

I would not worry about latency in this case unless I had a very specific use case.

|improve this answer
  • the distance between switches is between 50 and 100 meters and the longest chain will be of 6 switches. I was thinking if it is better each one to have an uplink to the core network. – vladiz May 10 '14 at 19:07
  • As I responded already, for that application and that distances you won't notice the difference. – Łukasz Bromirski May 10 '14 at 19:42
  • It could still make sense to connect them to the core. It depends on requirements for availability etc. If switch nr 2 in a chain fails then you have 5 switches offline. It's always a balance of cost, fiber availability and number of ports available. – Daniel Dib May 10 '14 at 20:30
  • 2
    It is also worth noting that a vast majority of switches now-a-days only support Store and Forward. – Ryan Foley May 10 '14 at 22:29
  • Most latencies with switches comes from buffers, not processing times or transmission speed. Cut-through is only possible if the destination links are not already used. – BatchyX May 11 '14 at 9:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.