Summary form only given. High performance computing clusters scaling out fact indicates manageability will become more important than ever. Over time, a computer center tends facilitate multiple management frameworks from vendors to remote manage generations of heterogeneous HPC clusters to complete one task. The heterogeneous and scaling out computing info structure made HPCC/grid administration even more challenging and time consuming than before. Management interoperability is usually compromised or absent due to the heterogeneous environment. In order to solve this problem for the long run and further reduce the total cost of ownership, industry is defining the systems management architecture for server hardware (SMASH) initiative. The SMASH initiative is a suite of specifications, which standardize management interfaces and remote management architecture for heterogeneous computing environments. The suite of specifications includes unified command line protocol, resource discovery, and resource addressing and data model profiles. SMASH not only addresses complicated administration challenges as well as enables hardware independent remote manageability plus computing info structure status/performance aware job scheduling schemes and as a result, will bring HPC clusters/grid utilization rates to an even higher level. This poster uses figures to illustrate the challenges, corresponding SMASH specifications and point out the potential research directions in supercomputing space over SMASH implementations

Summary form only given. The effect of Intel/spl reg/ hyper-threading (HT) technology on a system's performance varies according to the characteristics of the application running on the system and the configuration of the system. High performance computing (HPC) clusters introduce additional variables, such as the math libraries used in solving linear algebra equations that can affect the performance of scientific and engineering applications in particular. We study the effect of HT on MPI-based applications by varying the math library. We configure an Intel-based HPC cluster and used the High Performance Linpack (HPL) benchmark to study the performance characteristics without changing hardware or communication parameters, but by linking the application with different math libraries. What we have found is that, even though the application or hardware parameters remain the same, HT may help or hinder the overall performance of the cluster depending on the computational efficiency of the mathematical functions.

Linux clusters of commodity computer systems and interconnects have become the fastest growing choice for building cost-effective high-performance parallel computing systems. The Parallel Virtual File System (PVFS) could potentially fulfill the requirements of large I/O-intensive parallel applications. It provides a high-performance parallel file system by striping file data across multiple cluster nodes, called I/O nodes. Therefore, the choice of storage devices on I/O nodes is crucial to PVFS. We study the impact of software RAIDs and hardware RAIDs on the performance of PVFS when they are used on I/O nodes. We first establish a baseline performance of both RAIDs in a stand-alone configuration. We then present the performance of PVFS for a workload comprising concurrent reads and writes using ROMIO MPI-IO, and for the BTIO benchmark with a noncontiguous access pattern. We found that software RAIDs have a comparable performance to hardware RAIDs, except for write operations that require file synchronization.

This paper presents a performance study of two commodity clusters built from two models of Dell PowerEdge servers. Both clusters have eight servers interconnected by GigaNet for fast message passing and by Fast Ethernet for Network File System (NFS) traffic. The two server models are different in processors, level 2 (L2) cache, speed of front-side bus (FSB), chipsets and memory subsystem. They represent generic servers from two generations of Intel-based architecture. In this study, we use well-known benchmark programs to understand how they perform for computation-intensive applications. We first study their performance in stand-alone environment to unveil the performance characteristic of a compute node. We further explore their aggregated performance when they are used in a cluster environment. We are particularly interested in their scalability, per-processor performance degradation due to memory contention and inter-process communications and the correlation between results from different benchmark programs. We found that L2 cache and memory subsystem have significant impact on computation-intensive parallel applications such as the NAS Parallel Benchmark (NPB) programs. For configurations with a large number of processors (or multiple processors per compute node), some of NPB programs perform better on platform with larger global L2 cache, even though the platform has slower processors, FSB and memory components.

GigaNet and Myrinet are two of the leading interconnects for clusters of commodity computer systems. Both provide memory-protected user-level network interface access, and deliver low-latency and high-bandwidth communication to applications. GigaNet is a connection-oriented interconnect based on a hardware implementation of Virtual Interface (VI) Architecture and Asynchronous Transfer Mode (ATM) technologies. Myrinet is a connection-less interconnect which leverages packet switching technologies from experimental Massively Parallel Processors (MPP) networks. This paper investigates their architectural differences and evaluates their performance on two commodity clusters based on two generations of Symmetric Multiple Processors (SMP) servers. The performance measurements reported here suggest that the implementation of Message Passing Interface (MPI) significantly affects the cluster performance. Although MPICH-GM over Myrinet demonstrates lower latency with small messages, the polling-driven implementation of MPICH-GM often leads to tight synchronization between communication processes and higher CPU overhead.

Introduces the design, implementation and evaluation of a generic software architecture for on-demand video servers. We describe different key components for controlling the storage and network devices within the server. The interactive collaborations between these software components are also illustrated. The experimental results indicate a very promising direction in exploring the right combinations of these software components. The server is thus able to increase the number of concurrent video accesses with the same hardware configuration. For instance, with the right combinations, the system achieved about 80% of the storage system bandwidth of four disks, about 70% of the storage system bandwidth of six disks, and generally reached the maximal achieved SCSI bandwidth when eight disks are used over two SCSI buses. Our research and experimental results are based on video servers currently under construction across a variety of hardware platforms, including SMP, DMP and clusters of PCs or workstations. The most advanced prototype server is based on an SGI shared-memory multiprocessor with a mass storage system consisting of RAID-3 disk arrays. With all the enabling/management schemes, we were able to further investigate interesting research issues by considering the user's access profiles for taking advantage of popular video titles. The results were significant, with a range of 60% improvement given a 512 kByte block size. In addition to the experimental results, theoretical performance models were also developed that closely match to our collected experimental results.

To extend the widespread use of the high performance parallel interface (HIPPI) as a networking solution for high-speed communications, the 25-meter distance limitation must be solved. Three options available for alleviating the problem of distance limitation are serial-HIPPI, HIPPI/SONET mapping or HIPPI-ATM mapping, and IP routing. Serial-HIPPI, HIPPI/SONET mapping and HIPPI-ATM mapping provide extended HIPPI connectivities at the physical layer, while IP routing forwards data between the HIPPI networks and other networks at the network layer. We study two feasible solutions of this problem, HIPPI tunneling (HIPPI-ATM mapping) and IP routing. We compare these two schemes in terms of network connectivities, protocol overhead, and flow control. The performance evaluation of one implementation of HIPPI tunneling and IP routing is presented. The experimental performance suggests that a high degree of bandwidth utilization was achieved by both HIPPI tunneling and IP routing in this implementation.

Future video-on-demand (VOD) servers will need to support many existing and emerging video data types. These data types include 15-fps (frames per second) animation, 30-fps NTSC (National Television Systems Committee) quality video and 80-fps HDTV (High Definition Television) video. The different display speeds and frame sizes of these various video types impose a major constraint on the design of VOD storage systems. This paper presents the results of an experimental study, conducted on a Silicon Graphics Inc. Onyx computer system, that investigated the impact of these video types on the design of a VOD storage system. The key issues involved in supporting these different video types in a VOD environment are: (1) the video allocation method, and (2) the proper block size (a block is a basic unit of several contiguous video frames that will be accessed from several disks each time a request is made) to use for data striping and retrieval. Two allocation schemes, logical volume striping and application level striping, along with varying frame and block sizes for each of the three different video data types are examined. The focus of our study is to determine the maximum number of concurrent accesses that can be supported with a guaranteed quality of service. The degree of scalability (i.e., striping data over more disk arrays) of the experimental VOD system used is also studied.

Future video-on-demand (VoD) servers will need to support many existing and emerging video data types. These data types include 15-fps (frames per second) animation, 30-fps National Television Systems Committee (NTSC) quality video, and 60-fps high definition television (HDTV) video. The different display speeds and frame sizes of these various video types impose a major constraint on the design of VoD storage systems. This paper presents the results of an experimental study, conducted on a Silicon Graphics Inc. (SGI) Onyx computer system, that investigated the impact of these video types on the design of a VoD storage system. The key issues involved in supporting these different video types in a VoD environment are as follows: (1) the video allocation method and (2) the proper block size (a "block" is a basic unit of several contiguous video frames that will be accessed from several disks each time a request is made) to use for data striping and retrieval. Two allocation schemes, logical volume striping and application level striping, along with varying frame and block sizes for each of the three different video data types are examined. The focus of our study is to determine the maximum number of concurrent accesses that can be supported with a guaranteed quality-of-service (QoS). The degree of scalability (i.e., striping data over more disk arrays) of the experimental VoD system used is also studied. Based on our experimental results, application level striping demands smaller block sizes for all three video types, and more concurrent accesses can be distributed over the storage devices. The experimental results demonstrate that application level striping has excellent scalability for animation and NTSC videos.

Communication between processors has long been the bottleneck of distributed network computing. However, recent progress in switch-based high-speed local area networks (LANs) may be changing this situation. Asynchronous transfer mode (ATM) is one of the most widely-accepted and emerging high-speed network standards which can potentially satisfy the communication needs of distributed network computing. We investigate distributed network computing over local ATM networks. We first study the performance characteristics involving end-to-end communication in an environment that includes several types of workstations interconnected via a Fore Systems' ASX-100 ATM switch. We then compare the communication performance of four different application programming interfaces (APIs). The four APIs were Fore Systems' ATM API, the BSD socket programming interface, Sun's remote procedure call (RPC), and the parallel virtual machine (PVM) message passing library. Each API represents distributed programming at a different communication protocol layer. We evaluated two popular distributed applications, parallel matrix multiplication and parallel partial differential equations, over the local ATM network. The experimental results show that network computing is promising over local ATM networks, provided that the higher level protocols, device drivers, and network interfaces are improved.>

Emerging high-speed networks provide several hundred megabits per second to several gigabits per second of raw communication bandwidth. However, the maximum achievable throughput available to the end-user or application is quite limited. In order to fully utilize the network bandwidth and to improve the performance at the application level, a careful examination of I/O subsystems is essential. In this paper, we study one emerging high-speed network, the Fibre Channel network. The objectives of this study are: to understand how the I/O subsystem relates to network operations; to evaluate and analyze the performance of such a subsystem; and to propose possible approaches for improving the maximum achievable bandwidth. We show (by simply modifying device driver code) a 75% maximum achievable bandwidth improvement. Other ways of improving network performance are also discussed.>

Communication between processors has long been the bottleneck of distributed network computing. However recent progress in switch-based high-speed local area networks (LANs) may be changing the situation. Asynchronous transfer mode (ATM) is one of the most widely-accepted and emerging high-speed network standards which can potentially satisfy the communication needs of distributed network computing. We investigate distributed network computing over local ATM networks. We first study the performance characteristics involving end-to-end communication in an environment that includes several types of workstations interconnected via a Fore Systems' ASX-100 ATM Switch. We then compare the communication performance of four different application programming interfaces (APIs). The four APIs were Fore Systems ATM API, BSD socket programming interface, Sun's Remote Procedure Call (RPC), and the Parallel Virtual Machine (PVM) message passing library. Each API represents distributed programming at a different communication protocol layer. We evaluate parallel matrix multiplication over the local ATM network. The experimental results show that network computing is promising over local ATM networks.>