0

What happens with the network link between the PC and Server if the architecture is as follow: enter image description here

If the Client request UDP-traffic and the Server response will be 1Gbps UDP traffic. Would this result in lots of dropped packages at 100Mbps link due to the full buffer-memory at the switch?

If the Client uses TCP-traffic, all is fine due to the redundancy in the TCP-protocol?*

Can someone explain the details of what is actually going on (at the 100Mbps-link) when both Client and Server uses 1Gbps,but somewhere along the line, there is a 100Mbps link.

Improve this question
1
  • Never, ever do that. It is a very poor design that is fraught with single points of failure. Build trees, not loops or lines.
    – Ron Maupin ♦
    Commented Feb 12 at 19:04

2 Answers 2

Reset to default
2

Would this result in lots of dropped packages at 100Mbps link due to the full buffer-memory at the switch?

Yes. Note that not only packets carrying your UDP flow are dropped but any packets crossing the bottleneck may get lost.

If the Client uses TCP-traffic, all is fine due to the redundancy in the TCP-protocol?

I wouldn't call TCP redundant but it provides reliable delivery and - more importantly here - congestion control. TCP detects in-path bandwidth capabilities and problems, and throttles its traffic accordingly. That way, drops across the bottleneck may remain low, not only for the TCP stream in question but also for any other traffic.

Essentially, any application using UDP with significant bandwidths must implement its own congestion control.

If that bottleneck in your network is by design then it's a grave architectural problem.

Another significant problem in that network is the long chain of switches. Generally, you should create a tree of switches, not a chain: a 'core' switch in the center/root, and as many 'access' switches around it providing end-node connectivity as required.

If you can't get enough ports that way, you add an intermediate 'distribution' layer. A tree topology reduces the network diameter, the susceptibility for bottlenecks and it minimizes impact from link failure/degradation.

Improve this answer
1
  • 4 switches isn't a "long chain". (STP allows up to 7) If you look at the 100M link as the bridge between buildings, then it's 2 core switches, and 2 access switches. We don't know the physical nature of the network; don't read too much into it.
    – Ricky
    Commented Feb 12 at 18:12
-1

You would definitely see packets dropped.

The easiest & quickest fix is to upgrade the inter-swicth link or to bundle 8 interfaces together into LACP ( if you have free ports on both switches)

The inter-swicth link can take a lot of time depending on the infrastructure that you have. So LACP would be the easiset way to go aroud, as you would get atleast 800Mbps of link speed (something is better than nothing)

Improve this answer
2
  • 1
    Without questioning the reasons for the slow link and the chain topology, LAG is no good advice.
    – Zac67 ♦
    Commented Feb 12 at 10:53
  • 1
    ... and depending on the "load-distributability characteristics" of the given high volume UDP flow(s), you'd still not benefit from the 8 parallel links, and you'd possibly still heavily overload a subset of the 8 links.
    – Marc 'netztier' Luethi
    Commented Feb 12 at 14:27

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