Performance analysis of the Alpha 21364-based HP GS1280 multiprocessor

Abstract

This paper evaluates performance characteristics of the HP GS1280 shared memory multiprocessor system. The GS1280 system contains up to 64 Alpha 21364 CPUs connected together via a torus-based interconnect. We describe architectural features of the GS1280 system. We compare and contrast the GS1280 to the previousgeneration Alpha systems: AlphaServer GS320 and ES45/SC45. We further quantitatively show the performance effects of these features using application results and profiling data based on the built-in performance counters. We find that the HP GS1280 often provides 2 to 3 times the performance of the AlphaServer GS320 at similar clock frequencies. We find the key reasons for such performance gains are advances in memory, inter-processor, and I/O subsystem designs.

Regression Testing Based on UML Design Models

“…Each node contains detailed processor core, microcoded directory controller, memory controller, and DRAM models. Nodes communicate using a directory-based NACK-free, 3-hop cache-coherence protocol over an interconnect based on the HP GS1280 [3]. We list the relevant parameters in Table 1.…”

“…Spare memory nodes increase the cost of the system, however the overhead can be amortized over the size of the system. For typical DSM machines with four to sixteen memory nodes [3,10], the spare memory node overhead is lower than existing on-board redundancy mechanisms such as memory mirroring, RAID and DIMM sparing [2,8]. Data swap mode maintains both redundancy and performance by reducing the available system memory, while incurring no additional hardware costs.…”

Dense Gaussian Networks: Suitable Topologies for On-Chip Multiprocessors

“…It has been previously proved that reducing topological distances by skewing the wrap-around links in rectangular and "L-shape" Tori, results in better system performance (6)(7)(8) . Having lower distances implies higher network throughput and lower packet latencies which reduce the execution times of typical applications running over different kinds of multiprocessor platforms.…”

Adaptive History-Based Memory Schedulers

“…The Stream benchmarks, which others have used to measure the sustainable memory bandwidth of systems [6,13,3,16], consist of 4 simple vector kernels: Copy, Scale, Sum, and Triad. The NAS benchmarks are well known scientific benchmarks, which are fairly data-intensive.…”