It is important to establish the technology to accommodate best effort TCP/IP traffic over wide area ATM networks. The UBR (Unspecified Bit Rate) service category is the most typical service category for the best effort traffic, especially in the LAN environment. On the other hand, the VBR (Variable Bit Rate) service category with SCD (Selective Cell Discard) option is considered as the service category which is appropriate for wide area networks due to its fairness and minimum guarantee of the cell transmission using not only PCR (Peak Cell Rate) but SCR (Sustainable Cell Rate) and MBS (Maximum Burst Size). However, there is no actual evaluation for such service. We have, therefore, performed the experimental studies on TCP/IP over VBR with SCD along with UBR and VBR without SCD by VC (Virtual Channel) level policing when each TCP connection is mapped to a different VC. Through these experiments, we measured the link utilization of the effective data and the fairness between each obtained TCP throughput during the congestion of the ATM switch. From the results of the link utilization, the value is over 95% under the various conditions. Therefore, even in the case of the cell losses due to SCD or buffer overflow in ATM switch congestion, average throughput is almost the same as the value which equals the trunk line speed divided by the number of the accommodated TCP connections. From the results of the fairness, VBR with SCD per VC is better than UBR and also obtains better TCP throughput than VBR without SCD. Furthermore, to confirm those characteristics more generally, we adopt the accommodated TCP connections not only with the same TCP send/receive socket buffer size but with different sizes. Finally, we discuss the effectiveness between VBR with SCD and the other service categories, such as UBR and ABR (Available Bit Rate) and GFR (Guaranteed Frame Rate), and conclude that VBR with SCD is one of the most suitable ATM service categories for accommodating best effort traffic.

We study an asynchronous transfer mode (ATM) multiplexer with selective cell discarding (SCD), and propose as a new traffic parameter the ratio of high and low priority cell streams when a cell stream multiplexed is classified by the cell loss priority (CLP) field in a cell header. By having loss priority control, it is possible to increase the multiplexing gain. For performance analysis we assume that an on-of T bursty traffic is described with several traffic parameters and the proposed priority ratio, and is approximately modeled by a Markov-modulated deterministic process (MMDP). Assuming that several independent and homogeneous on-off bursty traffics with priority discrimination are multiplexed, we present an analytical procedure for the cell loss probability of each priority level in statistical cell multiplexing with loss priority control, and use the performance results for connection admission control (CAC). Also, we consider the effect of the proposed priority ratio. Although loss priority control increases the statistical multiplexing gain, it is not appropriate for the on-off bursty traffic to change the value of the high-priority ratio in order to obtain a larger multiplexing gain, since the admissible load is determined by the loss probability of low priority traffic for most cases and the values of the ratio in a certain range slightly affect it.