Allied Telesis Rapier Switch - Manual (Page 2)

2 Application Note
Allied Telesyn Guide to Ingress Rate Limiting on the Rapier-i
C613-43002-00 REV A
19 June 2003
Figure 1: The token bucket model
The consequences of dropped packets depend on the protocols involved. If the
application uses TCP at the Transport Layer, then dropped packets can have a
marked impact on throughput. If UDP is used at the Transport Layer, then the
consequences of dropped packets are normally handled at the Application
Layer. UDP is often used for real time applications where either the packet gets
through at the right time, or it doesn’t get through at all. Applications such as
voice suit this because the packet must get through at the right time; if it does
not, there is simply a glitch in the voice stream, provided that the majority of
the traffic is passed successfully.
The nature of Ingress Rate Limiting, and the way that TCP traffic responds to it,
combine to produce an apparent problem with throughput in TCP sessions. As
such it is not a fault. The Rapier-i switches provide Ingress Rate Limiting on a
per-port basis. It is not ‘TCP session bandwidth limiting’, or what is also called
‘traffic shaping’.
The connection-oriented nature of a TCP session, with its acknowledgements,
timout periods, resends, and sliding windows, means that dropped packets
have an impact on end-to-end throughput. As a consequence, the actual
measured throughput may be less than the configured Ingress Rate Limit.
The adverse effects on TCP traffic are more noticable when Ingress Rate
Limiting is set to low values such as 2Mbps. When rate limiting is set to
2Mbps, it takes less time for tokens to run out than when rate limiting is set to,
say, 50Mbps. This is because the rate of token replenishment is lower. TCP
timeout stand down periods will be frequent if the Ingress Rate Limit is set to a
rate that is a small fraction of the application’s sending rate, and the end-to-end
throughput will be lower than expected.
A solution for the Allied Telesyn Rapier-i
It is possible to use Egress Rate Limiting to simulate Ingress Rate Limiting with
the benefit of FIFO queueing using the following solution.
This solution requires 3 switch ports.
For this solution example, ports 1,2 and 3 and a VLAN named “two” are used.
The throughput is limited to 2Mbps.
The steps for the solution are:
■ Create a VLAN with two ports, using the command:
create vlan=two vid=2