Data Packet Loss in a Queue with Limited Buffer Space
Jose Gil, Motorola Research
A system that occurs in several contexts in computer and data networks has the generic form shown in Figure 1.
Figure 1: Diagram of system
There are N sources, and each source n transmits bursts of data at speed Rn packets a second, and they share a medium to which access is controlled by a queueing discipline. The queue has a limited buffer space B, and so in some conditions packets are dropped. The system feeds back to source n the proportion pn of its packets that were dropped, and the source then adjusts its sending rate according to the formula
Here K, C and D are to be treated as known dimensionless constants, and RTTn is the round-trip-time of the packets, from the time source n sends them to the time it receives the feedback that a proportion pn of them were dropped.
The problem is to find the equilibrium point (Rn(pn), pn,RTTn), and also information about the statistics of the distribution, eg the variance of the packet loss distribution.
There are 2 cases to consider:
Packet discard policies:
There are 2 packet discard policies that might be in operation:
The graphs of packet drop probability against buffer occupancy in these 2 cases are illustrated in Figure 2
Figure 2: Packet loss probability for the two packet discard policies
There are 3 queueing disciplines of interest:
The queueing patterns in the first 2 cases are illustrated in Figure 3
Figure 3: FO and round robin queueing systems
The service time per data packet is random, and for the purposes of this study it can be treated as exponentially distributed with a known mean service rate mu_n, different packets being independent.
Priorities for the Study Group:
The combinations of cases, in order of most interest, are
This page last modified
by C. Breward