EXLUDUS'S REPLICATOR SOFTWARE SPEEDS CLUSTER

eXludus Technologies Inc. today reported that its RepliCator software product accelerated run times more than ten-fold over NFS and five-fold over a dedicated cluster file system, on a cluster located at AMD headquarters in Sunnyvale, California. The tests, which measure data provisioning throughput and NCBI-BLAST performance, were run through multiple network interfaces and file serving devices on the Niobe reference cluster (dual-processor Opteron processors) in the Advanced Micro Devices (NYSE: AMD) Development Center. The comparison tests were first run at various node sizes on the full 5.4-gigabyte Human Genome database and measured 3 kilobases of queries, per compute node, in each case. "RepliCator can substantially boost productivity and ROI for popular cluster and grid computing resources by alleviating a major performance bottleneck," according to eXludus Technologies founder and CEO Benoît Marchand. "Clusters and grids no longer sit idle for significant periods while jobs wait for their data. While current jobs are running, RepliCator stages the data for subsequent jobs." • On the bandwidth test, at 32 nodes, aggregate bandwidth for RepliCator (1.0 GB/sec.) was double that of the dedicated cluster file system and 9 times NFS bandwidth. In moving to 128 nodes, RepliCator's bandwidth scaled linearly to more than 4 GB/sec., while bandwidth for the cluster file system remained at 555 MB/sec. NFS bandwidth was at most 115 MB/sec. for all cluster sizes; RepliCator was up to 35 times faster. • For the NCBI-BLAST test, on the full 5.4-gigabyte database, RepliCator's advantage was dramatic. RepliCator’s run time at 64 nodes (529 seconds) was 2.7 times faster than the cluster file system (1462 seconds) and more than 10 times faster than NFS (5752 seconds). At 128 nodes, RepliCator (529 seconds) was 5 times faster than the cluster file system (2869 seconds). The NFS server case could not be run at 128 nodes. eXludus also simulated results for cluster sizes as large as 512 nodes. On the simulated results, RepliCator scaled linearly even up to that system size, suggesting the relative performance advantage would continue to grow as compared to NFS or a cluster file system. “AMD recognizes the importance of effective data management to maximize cluster performance," said Terri Hall, Vice President, Software Alliances, AMD. “The results obtained on the RepliCator data transfer management software from eXludus highly complements the AMD64 architecture. Through continued investments in Direct Connect architecture and multi-core processors, AMD is providing processor level throughput to support solution providers like eXludus. We are also pleased the AMD Developer Center is helping AMD customers and partners develop solutions more quickly to achieve the strongest potential for success.” According to industry analyst firm IDC, clusters are the fastest-growing segment of the technical server market and in 2005 represented about half of that $9.2 billion market. A growing array of standards, software and partnerships promises to greatly expand the newer, closely related category of grid computing. RepliCator broadcasts data to all nodes in a cluster or grid, exploiting the improved performance and capacity of local disks. It especially benefits throughput-oriented workloads with moderate to high data rate, but is worthwhile for data rates as low as 250 MB per hour per CPU. In addition, RepliCator is asynchronous, fault-tolerant with respect to network and node failures, and interoperable with users' existing applications, workload managers and operating systems (including support for Linux 2.4 and above). The AMD Developer Center is designed to help AMD customers and partners develop their AMD64 technology-based solutions more quickly and efficiently and build in the strongest potential for success.