Escali 4.4 manual Communication protocols on DAT-devices, 6 SCI, Channel buffer

Page 28

Section: 2.3 Communication protocols on DAT-devices

2.2.6 SCI

This is a built-in device that uses the Scali SCI driver and library (ScaSCI). This driver is for the Dolphin SCI network cards. Please see the ScaSCI Release Notes for specific requirements. This device is straight forward and requires no configuration itself, but for multi-dimensional toruses (2D and 3D) the Scali SCI Management system (ScaConf) needs to be running somewhere in your system. Refer to Appendix C for installation and configuration of the Scali SCI Management software.

2.3 Communication protocols on DAT-devices

In SMC, the communication protocol used to transfer data between a sender and a receiver depends on the size of the message to transmit, as illustrated in Figure 2-3:.

Increasing message size

Transporter protocol: message size > eager_size

Eagerbuffering protocol:

channel_inline_threshold < message size <= eager_size

Inlining protocol

0 <= message size <= channel_inline_threshold

Figure 2-3:Thresholds for different communication protocol

The default thresholds that control whether a message belongs to the inlining, eagerbuffering or transporter protocols can be controlled from the application launch program (mpimon) described in chapter 3.

Figure 2-4:illustrates the node resources associated with communication and mechanisms implemented in Scali MPI Connect for handling messages of different sizes. The three communication protocols from Figure 2-3:rely on buffers located in the main memory of the nodes. This memory is allocated as shared, i.e., it is not private to a particular process in the node. Each process has one set of receiving buffers for of the processes it communicates with. As the figure shows all communication relies on the sending process depositing messages directly into the communication buffers of the receiver. For Inline and Eagerbuffering the management of the buffer resources does not require participation from the receiving process, because of their designs as ring buffers.

2.3.1 Channel buffer

The Channel ringbuffer is divided into equally sized entries. The size varies differs for different architectures and networks; see Scali MPI Connect Release Notes” for details. An entry in the ringbuffer, which is used to hold the information forming the message envelope, is reserved each time a message is being sent, and is used by the inline protocol, the eagerbuffering protocol, and the transporter protocol. In addition, one ore more entries are used by the inline protocol for application data being transmitted.

Scali MPI Connect Release 4.4 Users Guide

16

Image 28
Contents Scali MPI ConnectTM Users Guide Acknowledgement Copyright 1999-2005 Scali AS. All rights reservedScali Bronze Software Certificate II Software License Terms Commencement MaintenanceGrant of License Support License ManagerSub-license and distribution Export RequirementsSCALI’s Obligations LICENSEE’s ObligationsTitle to Intellectual Property Rights TransferWarranty of Title and Substantial Performance Compliance with LicensesLimitation on Remedies and Liabilities Scali MPI Connect Release 4.4 Users Guide ViiProprietary Information MiscellaneousGoverning Law Scali MPI Connect Release 4.4 Users Guide Table of contents Profiling with Scali MPI Connect Appendix a Example MPI code Scali MPI Connect Release 4.4 Users Guide Chapter Scali MPI Connect product contextScali mailing lists SMC FAQ SMC release documents Problem reportsSupport Platforms supportedHow to read this guide Acronyms and abbreviationsLicensing FeedbackNIC Typographic conventions Terms and conventionsGUI style font Typographic conventions Description of Scali MPI Connect Scali MPI Connect componentsSMC network devices Direct Access Transport DAT Network devicesShared Memory Device Ethernet DevicesUsing detstat Using detctl3.2 DET Myrinet Infiniband4.1 GM 5.1 IB6 SCI Communication protocols on DAT-devicesChannel buffer Eagerbuffering protocol Inlining protocolTransporter protocol Support for other interconnects MPI-2 FeaturesZerocopy protocol Scali MPI Connect Release 4.4 Users Guide MPI-2 Features Setting up a Scali MPI Connect environment Compiling and linkingScali MPI Connect environment variables RunningCompiler support Linker flagsRunning Scali MPI Connect programs Naming conventionsBasic usage Mpimon monitor programIdentity of parallel processes Controlling options to mpimon Standard inputStandard output Program specHow to provide options to mpimon Network optionsMpirun wrapper script Mpirun usageRunning with tcp error detection Tfdr Suspending and resuming jobsRunning with dynamic interconnect failover capabilities Part partDebugging and profiling Debugging with a sequential debuggerUsing built-in segment protect violation handler Built-in-tools for debuggingAssistance for external profiling Debugging with Etnus TotalviewCommunication resources on DAT-devices Controlling communication resourcesChannelinlinethreshold size to set threshold for inlining Using MPIIsend, MPIIrecv Using MPIBsendGood programming practice with SMC Matching MPIRecv with MPIProbeError and warning messages User interface errors and warningsFatal errors Unsafe MPI programsMpimon options Prefix Giving numeric values to mpimonPostfix Scali MPI Connect Release 4.4 Users Guide Profiling with Scali MPI Connect ExampleUsing Scali MPI Connect built-in trace TracingAbsRank MPIcallcommNamerankcall-dependant-parameters where +relSecs S eTime whereFeatures ExampleUsing Scali MPI Connect built-in timing TimingMPIcallDcallsDtimeDfreq TcallsTtimeTfreq Using the scanalyze Commrank recv from fromworldFromcommonFieldsCommrank send to toworldTocommonFields where Count!avrLen!zroLen!inline!eager!transporter! whereFor timing Using SMCs built-in CPU-usage functionality This produces the following reportScali MPI Connect Release 4.4 Users Guide Tuning communication resources Automatic buffer managementHow to optimize MPI performance BenchmarkingCaching the application program on the nodes First iteration is very slowCollective operations Memory consumption increase after warm-upFinding the best algorithm Programs in the ScaMPItst package Appendix aImage contrast enhancement Scali MPI Connect Release 4.4 Users Guide File format OriginalWhy does not my program start to run? When things do not work troubleshootingAppendix B Why does my program terminate abnormally? Why can I not start mpid?General problems Per node installation of Scali MPI Connect Appendix CInstall Scali MPI Connect for TCP/IP Install Scali MPI Connect for Direct EthernetInstall Scali MPI Connect for Myrinet ExampleInstall Scali MPI Connect for Infiniband Install Scali MPI Connect for SCIInstall and configure SCI management software License optionsTroubleshooting Network providers Uninstalling SMCScali kernel drivers Troubleshooting 3rdparty DAT providers Troubleshooting the GM providerScali MPI Connect Release 4.4 Users Guide Bracket expansion Appendix D Bracket expansion and groupingGrouping Scali MPI Connect Release 4.4 Users Guide Appendix E Related documentationScali MPI Connect Release 4.4 Users Guide List of figures Scali MPI Connect Release 4.4 Users Guide Index Transporter protocolSSP