Threshold-based filtering buffer management scheme in a shared buffer packet switch core part of the paper

TF cache management solution:

The TF scheme aims to use shared buffers efficiently and fairly, that is, to minimize the overall packet loss probability and guarantee fair buffer usage among different output ports.Output ports are classified as active or inactive output ports based on queue length, and the classification is based on a dedicated buffer allocation factor (B/N), where B is the total buffer size and N is the number of output ports in Definition 1.By reserving sufficient buffer size for all inactive output ports in overload mode in Definition 2, TF can protect arriving packets destined for inactive output ports from being dropped.

Definition 1: If the queue length of output port i is greater than B/N, output port i is called an active output port; otherwise, it is called an inactive output port.

Definition 2: If the total queue length is greater than B-T, the state of packet exchange is called overload mode; otherwise, it is called non-overload mode, where T is the TF threshold factor in this scheme.

The value of T should be kept as small as possible so that all output ports can share and use more buffer space.This can increase buffer utilization.However, a very small value of T will cause most incoming packets to inactive output ports to be dropped due to insufficient buffer space reserved.In this case, the buffering behavior of TF is similar to the drop-tail of CS, so the overall packet loss probability is greatly increased.Next, we show the operation of the TF scheme in non-reload mode.

Where Xi(t) is the maximum allowable buffer size for output port i at time t and Ai(t) is the number of all incoming packets destined for output port i at time t.When the switch state is in the non-overload mode, the operation of the TF is shown in Fig. 1(a).In Figure 1(a), all arriving packets are admitted into the buffer regardless of whether their destination is an active or inactive output port, where Q(t) is denoted as the total queue length at time t.Therefore, the maximum allowable buffer size for output port i at time t is equal to the number of all incoming packets destined for output port i at time t.Subsequently, (2) is used to show the operation of the TF scheme in reload mode.

where Sa(t) is the set of active output ports at time t and Sin(t) is the set of inactive output ports at time t.When the switch state is in overload mode, the operation of TF is shown in Fig. 1(b).In Figure 1(b), TF starts filtering arriving packets based on their destination.If the destination of incoming packets is an active output port, the filter simply blocks all incoming packets.On the other hand, if the destination of incoming packets is an inactive output port, then all those packets are allowed into the buffer.With a sufficient value of T, TF can guarantee that packets arriving at inactive output ports go into the buffer.This means that TF will not limit the length increase of inactive output ports unless they are diverted to active output ports.Therefore, TF is able to provide fairness and an overall low packet loss rate.