Network design aspects that influence cost and performance can be classified according to their distance from the applications, into issues concerning topology, switch technology, link technology, network adapter, and communication library. The network adapter has a privileged position to take decisions with more global information than any other component in the network. It receives feedback from the switches and requests from the communication libraries and applications. Also, compared to a network switch, an adapter has access to significantly more memory (host memory and on-chip memory) and memory bandwidth (which typically exceeds network bandwidth). The potential of the adapter to improve global network performance has not yet been fully exploited. In this work we show a series of noticeable performance improvements (of at least 10% to 15%) for medium-sized message exchanges in typical HPC communication patterns by optimizing message segmentation and packet injection policies, that can be implemented in an adapter's firmware inexpensively. We also show that implementing equivalent solutions in the switch (as opposed to the adapter) leads to only marginal performance improvements as the ones obtained by controlling the segmentation and injection policy at the adapter, while involving significantly more cost. In addition, enhancing the adapter will lead to less hardware complexity in the switches, thus reducing cost and energy consumption.

As processors become more powerful and clusters larger, users will exploit this increased power to progressively run larger and larger problems. Today's datasets in biology, physics or multimedia applications are huge and require high performance storage sub-systems. As a result, the hot spot of cluster computing is gradually moving from high performance computing to high performance IO. The solutions proposed by the parallel file-system community try to improve performance by working at the kernel level to enhance the regular IO design or by using a dedicated storage area network like Fiber Channel. We propose a new design to merge the communication network and the storage network at the best price. We have implemented it in OPIOM with the Myrinet interconnect. OPIOM moves data asynchronously from SCSI disks to the embedded memory of a Myrinet interface in order to send it to a remote node. This design presents attractive features: high performance and extremely low host overhead.

Article Free Access Share on BIP-SMP: high performance message passing over a cluster of commodity SMPs Authors: Patrick Geoffray RHDAC, project CNRS-INRIA ReMap, LHPC (Matra Systèmes & Information) ISTIL UCB-Lyon 69622 VILLEURBANNE RHDAC, project CNRS-INRIA ReMap, LHPC (Matra Systèmes & Information) ISTIL UCB-Lyon 69622 VILLEURBANNEView Profile , Loïc Prylli LIP, project CNRS-INRIA ReMap, LHPC (ENS-Lyon), ENS-Lyon, 69634 LYON LIP, project CNRS-INRIA ReMap, LHPC (ENS-Lyon), ENS-Lyon, 69634 LYONView Profile , Bernard Tourancheau RHDAC, project CNRS-INRIA ReMap, LHPC (Matra Systèmes & Information) ISTIL UCB-Lyon 69622 VILLEURBANNE and LIP, project CNRS-INRIA ReMap, LHPC (ENS-Lyon), ENS-Lyon, 69634 LYON RHDAC, project CNRS-INRIA ReMap, LHPC (Matra Systèmes & Information) ISTIL UCB-Lyon 69622 VILLEURBANNE and LIP, project CNRS-INRIA ReMap, LHPC (ENS-Lyon), ENS-Lyon, 69634 LYONView Profile Authors Info & Claims SC '99: Proceedings of the 1999 ACM/IEEE conference on SupercomputingJanuary 1999 Pages 20–eshttps://doi.org/10.1145/331532.331552Published:01 January 1999Publication History 23citation404DownloadsMetricsTotal Citations23Total Downloads404Last 12 Months21Last 6 weeks5 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteeReaderPDF