Steiner tree algorithms have been investigated for multicast services. The traffic on a multicast service is streamed from a server to multiple terminals, thus a directed Steiner tree should be used. Our proposed algorithm of this type, multiplex-aware route selection (MARS), produces the same tree as minimum-cost path heuristics (MPH). However, the computational complexity of MARS is O(mnlog(m)) compared with the O(m2n) of MPH, where m indicates the number of end nodes of the tree and n indicates the number of nodes on the network. We implemented the MARS algorithm and compared it with MPH and selective closest terminal first algorithm, which is an improvement of MPH, and also demonstrated the efficiency of MARS in terms of processing time and multicast tree cost.

We propose a new service for the L1VPN (layer-1 virtual private network), in which an L1VPN customer can manage and control its own L1VPN from an end-to-end point of view. In the service, a customer can change its routing policy on the basis of a decision and set network notification policy for individual VPN users. These operations are conducted by an L1VPN NMS (network management system), which is distributed online by an L1VPN provider in EJB (enterprise java beans) format. In addition to the L1VPN NMS, EJB-based customer domain NMSs that manage individual customer domains are also delivered to individual customers. In cooperation with the provider NMS, which is for the L1VPN provider network, and the customer domain NMSs, an L1VPN NMS can update the L1VPN logical information from provider and customer domains. The L1VPN NMS receives alarm notifications from both NMSs and forwards them to IP users who are affected by the notifications. We evaluate the effect of an L1VPN on alarm notification time because swift alarm notification is critical for IP users. In addition, we evaluate the effect of deploying multiple customer domain NMSs in one Linux NMS server.

The IPTV service, in which high-capacity content is broadcast from the IPTV server to a huge number of users, is becoming very popular. To establish an effective IPTV network, we need to minimize the cost of the IPTV multicast tree. That tree consists of a node, to which an IPTV server is connected, as the root node and other nodes, to which users are connected. In addition, we have to consider users who belong to multiple network domains. In this paper, we apply hierarchically distributed path computation elements (HDPCEs) to cooperatively create appropriate IPTV trees for multidomain users. Each HDPCE shares the burden of creating a multidomain multicast tree. Thus, we can reduce the computational burdens of creating a multicast tree. In addition, we enable choosing three different algorithms to create the cheapest multicast trees in individual domains. One of them is a new multiplex-aware-route- selection (MARS) algorithm. We evaluated the applicability of the three algorithms to various types of domains depending on network conditions.

Recently, requirements for integration of various networks, such as optical fiber networks and IP networks, are becoming stricter. To achieve this integration, generalized multiprotocol label switching (GMPLS) is now receiving more attention. The main feature of a GMPLS network is its multilayered complexity, which sometimes places a large burden on source GMPLS routers to determine the optimal routes to destinations and causes label switching path (LSP)-setup delays. To reduce this source-router burden, we propose a hierarchically distributed path computation server (HDPCS) that reduces the setup delay and conducts flexible routing depending on each GMPLS domain's or each GMPLS layer's routing policy. Because each HDPCS is allocated to each domain, including interdomains, which have several domains under them, each domain's security is also assured depending on each domain's security policy. Forwarding adjacency (FA) is introduced as one of the countermeasures against GMPLS multilayered routing problems and the OSPF routing protocol is applied to determine the end-to-end routes. Therefore, we compare the performance of the HDPCS with OSPF-based routing and demonstrate its applicability

A cell-discard method has been developed for use in networks supporting TCP-over-ATM traffic with an unspecified bit rate. In this "concentrated cell discard" (CCD) method, when an ATM switch becomes congested, consecutive cells are discarded from the front of the buffer, and cell discarding is suspended for a specified period of time. This method prevents TCP timeouts, which are the main reason for the decreasing TCP throughput in today's TCP-over-ATM networks, and shortens the duration of the congestion by avoiding the long delay experienced with the TCP fast-retransmission algorithm before the sender's TCP receives duplicate acknowledgments. Even though CCD does not consider TCP packet size or boundaries, the throughput with CCD was found by computer simulation to actually be higher than with early packet discard.

The service specific connection oriented protocol is now being studied for use as the ATM adaptation layer protocol for connection oriented data communication and UNI/NNI signaling in the B-ISDN. The key feature of this protocol is a numbered poll mechanism that enables status information to be exchanged frequently and that is helpful for selective retransmission for frame loss. This paper examines this mechanism from the viewpoints of polling interval and frame length. Because this protocol was designed to perform selective retransmission for error recovery, some requirements, such as saving buffer, are not met. Two enhancements are therefore proposed: one consists of a Go-back-N-like procedure and nonassured credit, and the other is an SD PDU with a P bit. Guidelines to determine the POLL PDU interval are given.>