We developed a SPARC-V9 processor, the SPARC64 V. It has an operating frequency of 1.35 GHz and contains 191 million transistors fabricated using 0.13-µm CMOS technology with eight-layer copper metallization. SPECjbb2000 (CPU# 32) is 492683, highest on the market and 42% higher than the next highest system. SPEC CPU2000 performance is 858 for SPECint and 1228 for SPECfp. The processor is designed to provide the high system performance and high reliability required of enterprise server systems. It is also designed to address the performance requirements of high-performance computing. During our development of several generations of mainframe processors, we conducted many related experiments, and obtained enterprise server system (EPS) development skills, an understanding of EPS workload characteristics, and technology that provides high reliability, availability, and serviceability. We used those as bases of the new processor development. The approach quite effectively moves beyond differences between mainframe and SPARC systems. At the beginning of development and before the start of hardware design, we developed a software performance simulator so we could understand the performance impacts of created specifications, thereby enabling us to make appropriate decisions about hardware design. We took this approach to solve performance problems before tape-out and avoid spending additional time on design update and physical machine reconstruction. We were successful, completing the high-performance processor development on schedule and in a short time. This paper describes the SPARC64 V microprocessor and performance analyses for development of its design.

The bandwidth explosion ushered in by the popularity of the Internet has spurred the recent acceleration in the development and deployment of equipment supporting packet based broadband services. This coupled with the widespread deployment of WDM based Optical Transport Systems in the core network to satisfy the corresponding increase in capacity demand, has led network planners for tighter coordination between IP and Optical layers to increase reliability, robustness of next generation backbone network. In this paper, we propose a solution known as border model, which is tailored to address deployment concerns associated with GMPLS technology in existing networks. We extend our proposal to include, "Border model based Multi-layer service network architecture," to provide coordinated multi-layer IP and Optical services, for different network design scenarios. Resource Control is an important aspect of multi-layer service networks. This paper examines next generation requirements for resource control, defines resource control architecture and presents some evaluation results for multi-layer recovery techniques in the context of Multi-layer service network based on border model.

Implementing the fast-responding multi-layer service network (MLSN) functionality will allow the IP/MPLS service network logical topology and Optical Virtual Network topology to be reconfigured dynamically according to the traffic pattern on the network. Direct links can be created or removed in the logical IP/MPLS service network topology, when either extra capacity in MLSN core is needed or existing capacity in core is no longer required. Reconfiguring the logical and virtual network topologies constitute a new manner by which Traffic Engineering (TE) can solve or avoid network congestion problems and service degradations. As both IP and optical network layers are involved, this is called Multi-layer Traffic Engineering. We proposed border model based MLSN architecture in [5]. In this paper, we define the realization of Multi-Layer TE functions using Path Computation Element (PCE) for Border model based MLSN. It defines nodal requirements for multi-layer TE. Requirements of communication protocol between PCC (Path Computation Client) and PCE is introduced. It presents Virtual Network Topology (VNT) scenarios and steps involved along with examples for PCE-based VNT reconfiguration triggered by network failure, where VNT is a set of different layer's network resource accumulation.

Broadcast and synchronization techniques are used for cache coherence control in conventional larger scale snoop-based SMP systems. The penalty for synchronization is directly proportional to system size. Meanwhile, advances in LSI technology now enable placing a memory controller on a CPU die. The latency to access directly linked memory is drastically reduced by an on-die controller. Developing an enterprise server system with these CPUs allows us an opportunity to achieve higher performance. Though the penalty of synchronization is counted whenever a cache miss occurs, it is necessary to improve the coherence method to receive the full benefit of this effect. In this paper, we demonstrate a coherence directory organization that fits into DSM enterprise server systems. Originally, a directory-based method was adopted in high performance computing systems because of its huge scalability in comparison with snoop-based method. Though directory capacity miss and long directory access latency are the major problems of this method, the relaxed scalability requirement of enterprise servers is advantageous to us to solve these problems along with an advanced LSI technology. Our proposed directory solves both problems by implementing a full bit vector level map of the coherence directory on an LSI chip. Our experimental results validate that a system controlled by our proposed directory can surpass a snoop-based system in performance even without applying data localization optimization to an online transaction processing (OLTP) workload.

The rapidly increasing bandwidth requirements of IP traffic mean that networks based on optical technologies in conjunction with IP routing technologies will provide the backbone of the next generation Internet. One of the major issues is how to construct an optical-technology-based backbone network that offers the economical transport of large-scale IP/MPLS services while achieving reliable, robust network. The key to achieving this objective lies in multilayer coordination technologies using Multi-Layer Service Network [MLSN] Architecture, that we previously proposed [2]. One of the important aspects of MLSN architecture is ability to effectively use GMPLS network resources by IP/MPLS service networks. We propose extensions to previously proposed MLSN architecture. The proposed extensions to MLSN architecture are tailored to address "service virtualization and separation" of various service networks over GMPLS backbone. As a part of this extended MLSN architecture, we introduce novel concepts known as Logical Router (LR) and Virtual Router (VR) that would enable border router to be services domain router, so that it can connect multiple service networks such as L2VPN, L3VPN etc., over GMPLS backbone by offering service separation or virtualization. This service separation/isolation greatly enhances the reliability of next generation networks, as any failure on one service should be isolated from others. We evaluate our extended network architecture against requirements for the large scale network targeting at introducing such new technology to cope with vast traffic explosion and challenges in operation and service provision sophistication.

This paper proposes a network design scheme for Virtual Private Network (VPN) services. Traditionally, network design to compute required amount of resource is based on static point-to-point resource demand. This scheme is effective for traditional private line services. However, since VPN services allow multi-site connectivity for customers, it may not be appropriate to design a network based on static point-to-point resource demand. In particular, this scheme is not effective when the traffic pattern changes over time. Therefore, network design for VPN services introduces a new challenge in order to comply with traffic flexibility. There are conventional studies tackling this issue. In those studies, by defining a resource demand model considering flexibility, and designing the network based on this model, amount of resource required can be computed. However, there are some deficiencies in those studies. This paper proposes a new network design scheme, consisting of two components. The first one is a new resource demand model, created by extending conventional resource demand models, that can specify resource demand more precisely. The second one is a new network design algorithm for this resource demand model. Simulations are conducted to evaluate the performance of the proposed network design scheme, and the results show significant performance improvement against conventional schemes. In addition, deployment considerations of the proposed scheme are analyzed.

This paper proposes resource management in Layer 1 Virtual Private Networks (VPNs). We have been proposing Layer 1 VPNs that provide layer 1 services to multiple customers over the single optical network with per VPN control and management capabilities. We have proposed two resource management models for Layer 1 VPNs, which constitute different class of services. One is the shared model, where resources are shared among VPNs. The other is the dedicated model, where resources are explicitly pre-assigned to each VPN. In this paper, after introducing an overview of Layer 1 VPNs, we evaluate several path computation algorithms for these two models focusing on the multi layer network scenario. In the shared model, there are several existing studies for non-VPN cases, but considerations for VPN cases are not investigated. This paper evaluates algorithms originally proposed for non-VPN cases for use in VPN cases. Simulation results show that the path computation algorithm that works as saving layer 1 resources achieves better resource sharing effect. In the dedicated model, the problem is identical to non-VPN cases. There is one conventional algorithm, but amount of available resources is not well considered. We propose a novel path computation algorithm. Simulation results show effectiveness of our proposed algorithm against the conventional algorithm. Furthermore, resource usage efficiency of two resource management models is compared. We analyze and propose applicability of resource management models.

This paper proposes the Path Computation Element (PCE)-based backbone network architecture and verifies its feasibility through implementation and experiments. PCE communication Protocol (PCEP) is implemented for communication between the PCE and the management system to control and manage Generalized Multi-Protocol Label Switching (GMPLS)-based backbone networks.