0

Sorry for my English.

Let's say there is a chat server in AWS and it has bandwidth limit like 5Gbps. Someday, massive user comes and it spikes a lot of traffic in a second and finally it overs 5Gbps in 15 seconds.

I'm curious what is gonna happen? does OS drop the inbound traffic? or traffics are not handled(pass to process) but stored some kind of queue internally in OS?

I'm asking this because I don't understand why there is no limitation of bandwidth or traffic in AWS ELB.

| improve this question | |
  • The traffic may be queued or dropped, depending on the configuration of the server and the network components that it connects to. But questions about server configuration are off topic here and should be asked on Server Fault. – Ron Trunk Aug 11 at 11:43
3

If the bandwidth limit is a physical link capacity, excess traffic is dropped. A very small amount might get queued and dequeued again when the bandwidth has decreased below the physical limit, but anything seriously exceeding the link capacity cannot be forwarded and is lost.

If it's a software limit, the exact handling depends on the configuration and the implementation. It's not uncommon to accept and forward short bursts exceeding the agreed bandwidth - "short" being anything between a few seconds and several minutes or even hours/days, as long as the average rate doesn't exceed said maximum.

Note that this is how network devices handle limits. Host configurations and behavior are explicitly off-topic here as are (cloud) networks not under your complete control.

| improve this answer | |
  • It's worth adding that when you're getting too much traffic and packets get dropped, the target system may or may not receive ECN notification about the fact. In worst case TCP/IP handles the missing packets by hopefully getting some packets through and requesting retry from the origin host for the missing packets in between. In case the end of stream is dropped without ECN support, the problem is noticed only after a timeout. A well designed congestion handling during routing should not drop last packet of a stream if it has any other choice. – Mikko Rantalainen Aug 11 at 14:01
2

It's all depends upon network connectivity bandwidth and server NIC card bandwidth threshold . If excess of traffic come on physical links then packets are queued resulting in congestion of physical links leads to packet drops and applications will experience latency in accessing .

Traffic is even get queued when server nic cards exceed its handling threshold range . So always ensure load balances traffic or use high bandwidth physical links as per your usages and use high bandwidth upgrade Server NIC cards to ensure and avoid congestion in traffic flow.

| improve this answer | |

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