Intel Manycore Platform Software Stack (Intel MPSS)

Transcription

1 Intel Manycore Platform Software Stack (Intel MPSS) Copyright Intel Corporation All Rights Reserved US Revision: World Wide Web:

2 Disclaimer and Legal Information You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein. No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest Intel product specifications and roadmaps. The products described may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request. Copies of documents which have an order number and are referenced in this document may be obtained by calling or by visiting: Intel, the Intel Logo, the Intel Inside logo, Xeon, Intel Xeon Phi, Intel Vtune are trademarks or registered trademarks of intel corporation or its subsidiaries in the U. S. and/or other countries. *other names and brands may be claimed as the property of others. Copyright 2017, Intel Corporation. All rights reserved. 2

3 Revision History Revision Number Description Revision Date Document update for the release of the Intel MPSS March Document update for the release of the Intel MPSS February Document update for the release of the Intel MPSS Updated the guide s structure to be more parallel with the user s guide for Linux*. Updated information about systools. January Document update for the release of the Intel MPSS November Document update for the release of the Intel MPSS Added instructions on how to establish bridged network connections with coprocessors, updated SSH access instructions, and supplemented Appendix B with description of the Intel MPSS System Tools Document update for the release of the Intel MPSS Added information on new configuration parameters Document update for the release of the Intel MPSS Improved document structure, new Appendix A with coprocessor configuration parameters, updated configuration instructions reflecting changes in the structure of configuration files. September 2016 September 2016 August Document update for the release of the Intel MPSS June Release of the Intel MPSS User s Guide for Windows* April

4 Table of Contents 1 About this manual Overview of this document Notational conventions Command syntax Terminology Intel MPSS at a glance Coprocessor hardware and system architecture Programming models Offload programming model Symmetric programming model Native programming model Intel MPSS software architecture and components The coprocessor operating system Intel MPSS middleware libraries Intel MPSS drivers and services Tools and utilities Related documentation Installation process overview Hardware and software prerequisites BIOS configuration Supported host operating systems Software prerequisites Installing Intel MPSS Intel MPSS installation Upgrade instructions Installing Windows* cross-sdk Uninstalling Intel MPSS Updating coprocessor s firmware Initial coprocessor boot Starting the coprocessor Initial configuration Specifying administrator s SSH key Validating Intel MPSS installation Coprocessor OS configuration Configuration files Updating configuration files Boot configuration Specifying the Linux* kernel Coprocessor-side kernel command line parameters Coprocessor s root file system Automatic boot Basic administration tasks Adding groups and users to the coprocessor s file system Removing users and groups from the coprocessor s file system Specifying the host SSH keys Troubleshooting

5 5 Network Configuration MAC address assignment Host firewall configuration Supported network configurations Static pair Bridging Interacting with the coprocessor External tools SSH access to the coprocessor Transferring files to and from the coprocessor Managing the coprocessor s file system Adding persistent files to the coprocessor s file system Using the smart package manager Editing files interactively Transferring files over SSH Installing RPM packages (root only) Virtual block devices Host file system share Using the server manager Using PowerShell* cmdlets to create an NFS share (optional) CIFS share on Windows* Configuring a CIFS share Mounting a temporary CIFS share Troubleshooting Intel MPSS component configuration and tuning Coprocessor operating system configuration Disabling and enabling the memory control group (cgroup) Enabling Windows* mic gdb debugging for offload processes Enabling Windows* mic debugging for myo applications COI configuration Limiting maximum number of COI processes COI offload user options Micuser ownership Intel MPSS tools micctrl Checking the coprocessor s state Booting the coprocessor Starting the coprocessor Shutting down the coprocessor Stopping the coprocessor Resetting the coprocessor Rebooting the coprocessor Waiting for the coprocessor state change Restoring the default PFS image Restoring the coprocessor default configuration Downloading the coprocessor kernel log Displaying the MPSS log micinfo Options

6 8.2.2 Usage miccheck Options Examples micsmc-cli Options Usage Examples micfw Options Usage Examples micflash Options Usage Examples micbios Options Usage The libsystools library Function groups A Intel MPSS configuration parameters A.1 Meta configuration A.1.1 Including other configuration files A.2 Boot control A.2.1 EfiImage A.2.2 KernelImage A.2.3 KernelSymbols A.2.4 InitRamFsImage A.2.5 AutoBoot A.3 File system configuration parameters A.3.1 RootFsImage A.3.2 RepoFsImage A.3.3 BlockDevice A.4 Kernel configuration A.4.1 KernelExtraCommandLine A.5 Network configuration A.5.1 Host network configuration A.5.2 A.5.3 Coprocessor network configuration MAC address assignment B Intel MPSS boot process B.1 Kernel command line B.2 Instruct the driver to boot the coprocessor B.3 Coprocessor Linux* kernel execution B.4 Coprocessor s root file system

7 List of Figures Figure 1 Example workstation configuration Figure 2 Intel Xeon Phi coprocessor x200 architecture Figure 3 Spectrum of programming models Figure 4 Static pair configuration Figure 5 Internal bridge configuration Figure 6 External bridge configuration Figure 7 Persistent file system concept Figure 8 Boot process for Intel MPSS List of Tables Table 1 Supported host operating systems

8 About this manual 1 About this manual This guide is intended to provide you with a firm understanding of the Intel Manycore Platform Software Stack (Intel MPSS), what it is, how to configure it, and how to use its components. This document pertains to systems containing at least one Intel Xeon Phi coprocessor x200. Please note that utilizing systems with both Intel Xeon Phi coprocessor x200 and previous generation coprocessors is not supported. 1.1 Overview of this document Section 2 provides a high level overview of the Intel Xeon Phi coprocessor x200 design and Intel MPSS architecture. Section 3 is a step-by-step guide to installing Intel MPSS. It also shows basic configuration steps. Section 4 shows how to configure the coprocessor OS Section 5 describes supported network configurations, explains when each might be used, and shows how to configure them. Additionally it shows how to interact with the coprocessor using a selection of external tools. Section 6 presents methods for managing the coprocessor s file systems, describes how to add supplementary software to the coprocessor OS, and shows how to configure and mount NFS and CIFS shares on the coprocessor. Section 7 presents configuration options for Intel MPSS components, including the coprocessor s Linux* kernel, and the COI offload interface Section 8 lists tools distributed with the software stack. Appendix A describes the Intel MPSS-specific configuration parameters. Appendix B is an in-depth description of the coprocessors boot process. 1.2 Notational conventions This document uses the following notational conventions. micctrl --start C:\Program Files\Intel\MPSS\ Commands and their arguments in prose sections are italicized. Files and directories in prose sections are italicized. 8

9 About this manual micn %INTEL_MPSS_HOME% COURIER text Italic COURIER text User> Admin> [micn]$ [micn]# Indicates any coprocessor name mic0, mic1, mic2 etc. where N=0, 1, (e. g. the micn.conf file resembles any of the files mic0.conf, mic1.conf etc.). Indicates the software stack installation folder. By default it is C:\Program Files\Intel\MPSS Indicates code and commands entered by the user. A backslash \ indicates the command continues into the next line. OS terminal output. Command entered on the host with user privileges. Command entered on the host with administrator privileges. Command entered on a coprocessor with user or root privileges. Command entered on a coprocessor with root privileges Command syntax The following conventions are used in this document: < > indicates a variable value to be supplied. [ ] indicates an optional component. (x y z) indicates a choice of values. Example syntax of the micctrl --start command: Admin> micctrl -start [mic0 micn] The optional [mic0 micn] parameter indicates a space-separated list of coprocessors, for example mic0 mic2 mic3 etc. 1.3 Terminology ABI Application Binary Interface CCL Coprocessor Communication Link 9

10 About this manual COI Coprocessor Offload Infrastructure Coprocessor Intel Xeon Phi coprocessor x200 DHCP Dynamic Host Configuration Protocol GDB GNU Debugger HCA Host Channel Adapter IPoIB Internet Protocol over InfiniBand* K1OM Architecture of the Intel Xeon Phi Coprocessor x100 Product Family LDAP Lightweight Directory Access Protocol Lustre A parallel, distributed file system MAC Media Access Control MIC Many Integrated Core MPI Message Passing Interface MPSS Intel Manycore Platform Software stack MYO Mine, Yours, Ours shared memory infrastructure NIS Network Information System OFED Open Fabric Enterprise Distribution PCIe3 PCI Express Gen 3.0 QPI Intel QuickPath Interconnect RHEL* Red Hat* Enterprise Linux* RPM RPM Package Manager 10

11 About this manual SCIF Symmetric Communication Interface SLES* SUSE* Linux* Enterprise Server SMP Symmetric Multi-Processor SSD Solid State Drive SSH Secure Shell Sysfs A virtual file system VEth Virtual Ethernet X86_64 Architecture of the Intel Xeon Phi Coprocessor x200 product family 11

12 Intel MPSS at a glance 2 Intel MPSS at a glance This section provides an overview of the Intel Manycore Platform Software Stack. It contains a high level description of Intel Xeon Phi coprocessor x200 hardware and system architecture, discusses programming models Intel MPSS is designed to support, shows how various components support those programming models, and concludes with a listing of other available documentation. 2.1 Coprocessor hardware and system architecture The coprocessor is a PCIe add-in card designed for installation into an Intel Xeon processor-based platform. A typical platform configuration is shown in Figure 1. Figure 1 Example workstation configuration The coprocessor is composed of up to 72 cores, each supporting 4 hardware threads and standard AVX 512 instructions. The coprocessor is equipped with up to 16GB of on-package, high-bandwidth MCDRAM memory (there is no traditional DDR memory). The coprocessor has a PCIe interface working as a Gen3 PCIe root port. A Non- Transparent Bridge (NTB) chip is used to connect the device as a PCIe end port to the host. 12

13 Intel MPSS at a glance The coprocessor has no permanent file system. Instead, the file system is implemented on the virtual block device located in the host s file system. Each coprocessor runs a standard Linux* kernel with some minor accommodations for the MIC hardware architecture. Because it runs its own OS, the coprocessor is not cache coherent with the host or other PCIe devices. Figure 2 Intel Xeon Phi coprocessor x200 architecture 13

14 Intel MPSS at a glance 2.2 Programming models The Offload, Symmetric and Native programming models offer a diverse range of usage models. An overview of these options is depicted in Figure 3. Figure 3 Spectrum of programming models Offload programming model In the Offload model, one or more processes of an application are launched on one or more host processors. These processes, represented in the figure by main(), can offload computation, represented by work(), to attached coprocessors, taking advantage of the many-core architecture with its wide vector units and high memory bandwidth. If the application is composed of more than one process, the processes often communicate using some form of message passing, such as MPI (Message Passing Interface) thus MPI_*() is shown on the host. This offload process is programmed via the use of offload pragmas supported by the Intel C/C++ and FORTRAN compilers. When an application is created with one of these compilers, offloaded execution will fall back to the host in the event that a coprocessor is not available. This is why an instance of work() is shown on the host as well Symmetric programming model The Symmetric programming model is convenient for existing HPC applications that are composed of multiple processes, each of which could run on the host or the coprocessor, and use some standard communication mechanism such as MPI. In this model, computation is not offloaded but rather remains within each of the processes comprising the application. In such cases, where the application is MPI based, OFED enables high bandwidth/low latency communication using installed Intel True Scale or Mellanox* InfiniBand* Host Channel Adapters. 14

15 Intel MPSS at a glance Native programming model The Native programming model is a variant of the symmetric model where one or more processes of an application are launched exclusively on one or more coprocessors. From the Intel MPSS architecture perspective, these programming models typically depend on SCIF and the Virtual Ethernet (VEth) driver to launch processes on the coprocessor(s). 2.3 Intel MPSS software architecture and components The coprocessor operating system Underlying all computation on the coprocessor is a standard Linux* kernel with changes in the Power Management (PM) driver and Performance Monitoring Unit (PMU) driver to support the coprocessor. The Linux* kernel and initial file system image are installed into the host file system as part of the software stack. After the installation each coprocessor will need to be configured in accordance with the expected workload/application. This configuration is covered in Section 4. The Linux* environment on the coprocessor utilizes BusyBox to provide a number of Linux* utilities. However, these utilities may have limited functionality when compared to similar tools provided with the typical Linux* distribution. For more information regarding BusyBox, refer to Intel MPSS middleware libraries The compiler runtimes depend on the COI (Coprocessor Offload Infrastructure) library to offload executables and data for execution on the coprocessor, and use MYO (Mine, Yours, Ours shared memory infrastructure) to provide a virtual shared memory model simplifying data sharing between processes on the host and coprocessor. Some functions in the Intel Math Kernel Library (Intel MKL) automatically offload work to the coprocessor using the COI library. COI, MYO, and other software stack components rely on the SCIF (Symmetric Communication Interface) user mode API for PCIe communication services between the host processor, coprocessors, and optionally installed InfiniBand* host channel adapters. SCIF delivers high bandwidth data transfers and sub-sec write latency to memory shared across PCIe, while abstracting the PCIe communication details. The COI, MYO, and SCIF libraries are also available to other applications. Section 2.4 lists additional documentation on these libraries Intel MPSS drivers and services The following drivers run on the host system enabling communication with the coprocessor: Mic.sys MicVeth.sys 15

16 Intel MPSS at a glance Drivers are responsible for initializing, booting, and managing the coprocessor. They also implement the host-side parts of the SCIF protocol, and Virtual Ethernet (Veth) - two communication interfaces between the host and coprocessor. The virtio block device (virtblk) uses the virtio data transfer mechanism to implement a block device on the coprocessor. The device saves data in a specified storage location on the host and can therefore persist across coprocessor reboots. An MpssControlService runs on the host, controlling access to the coprocessor s file system. The Intel Omni-Path Architecture enables direct data transfers between the coprocessor s memory and installed Intel Omni-Path Host Fabric Adapters. The ibscif (InfiniBand* over SCIF) driver emulates an InfiniBand* HCA to the higher levels of the OFED stack. This driver uses SCIF to provide high bandwidth, low latency communication between multiple coprocessors in a host platform, for example between MPI ranks on separate coprocessors Tools and utilities Intel MPSS includes several system management tools and utilities (for detailed instructions on utilizing these tools refer to Section 8): micctrl is a utility for controlling (boot, shutdown, reset) each (or all) of the installed coprocessors. micctrl also offers options to simplify the configuration process. Configuration tasks include controlling user access and adding coprocessors to a TCP/IP network. For detailed information execute: User> micctrl -h micinfo displays information about installed coprocessors, the host s operating system and the Intel MPSS host driver. For detailed information execute: User> micinfo -h micfw is used for updating the coprocessor s firmware and displaying the version of firmware loaded on the coprocessor or residing in a firmware image file. For detailed information about micfw execute: User> micfw -h miccheck runs a suite of sanity checks on host systems containing coprocessors. The win32com Python extension must be installed before running miccheck. For detailed information execute: User> miccheck --help micbios allows to change certain coprocessor s BIOS settings, as well as to use the Firmware Update Protection feature. User> micbios --help libsystools is a C-library interface to the coprocessor for applications running on the host. These are wrappers built on top of the MIC host driver interfaces, available in the form of WMI entries and scif calls. Various functions are available to query the 16

17 Intel MPSS at a glance state of each coprocessor, and list or modify some of its parameters. Library functions are categorized in the following groups: Identify Available Coprocessors. Error Reporting. Device Access. Query Coprocessor State. General Device Information. PCI Configuration Information. Memory Information. Processor Information. Coprocessor Core Information. Version Information. LED Mode Information. Turbo State Information. SMC Configuration Information. Throttle State Information. Coprocessor OS Power Management Configuration. Thermal Information. Power Utilization Information. Memory Utilization Information. Power Limits. For detailed information about libsystools, refer to the help page: C:\Program Files\Intel\MPSS\docs\libsystools\index.html. 2.4 Related documentation Listed below are documents installed alongside the software stack in their default locations. COI documentation and tutorials: The following COI documentation is installed during base Intel MPSS installation: C:\Program Files\Intel\MPSS\docs\coi\release_notes.txt C:\Program Files\Intel\MPSS\docs\coi\MIC_COI_API_Reference_Manual_1_0.pdf 17

18 Intel MPSS at a glance C:\Program Files\Intel\MPSS\sdk\include\intel-coi header files containing full API description C:\Program Files\Intel\MPSS\sdk\tutorials\coi\README_Windows_1.txt instructions on how to get coprocessor side binaries and general building and use of COI tutorials. MYO documentation and tutorials: MYO tutorials and other documentation are installed in the C:\Program Files\Intel\MPSS\sdk\tutorials\myo directory. The README.txt file in this directory contains instructions on running the MYO tutorials. SCIF documentation and tutorials: The C:\Program Files\Intel\MPSS\sdk\tutorials\scif\README.txt contains instructions on running the SCIF tutorials. 18

19 Installation process overview 3 Installation process overview This section describes how to install the coprocessor hardware and software and initialize the default configuration. Read through this section before proceeding with installation, to ensure the required components and facilities are available. To avoid setup issues follow these instructions in their presented order. 3.1 Hardware and software prerequisites Host system hardware Each coprocessor requires a 64-bit x16 PCIe slot with 75W power output capability. Refer to your motherboard s manual to identify compatible slots. Note: Installing more than 8 coprocessors in a single host platform is not supported. Note: Host platforms with more than 256GB are not supported yet. This limitation will be removed in the upcoming release. Coprocessor physical installation If PCIe devices are installed into IO hubs of different CPU sockets, they will communicate across the Quick Path Interconnect (QPI). The bandwidth of this communication is typically lower than communication bandwidth between two devices on the same IO hub BIOS configuration 1. Update your host platform s BIOS to the latest version and reset the settings to their default values. 2. Enable Large Base Address Registers (BAR) support in the host platform BIOS. BIOS and OS support for large (32+GB) Memory Mapped I/O Base Address Registers (MMIO BARs) above the 4GB address limit must be enabled. 3. Modify the following BIOS settings to ensure maximum and stable performance: a. If available on your platform, set the Extended ATR setting to 0x1. b. Enable Intel Turbo Boost Technology. 4. Disable Virtualization Technology for Directed I/O (VT-d). 19

20 Installation process overview Supported host operating systems Intel MPSS was validated against specific versions of Microsoft* Windows* as the host operating system. Table 1 lists supported host operating systems. Table 1 Supported host operating systems Supported Host OS Version Microsoft* Windows* 8.1 Microsoft* Windows* 10 Microsoft* Windows* Server 2012 R2 Microsoft* Windows* Server Software prerequisites The software stack requires the following software to be installed on the host system: Microsoft*.NET Framework* 4.5 or higher Python* x86-64 or higher (Python 3.x is not supported) Pywin32 build 220 or higher ( 3.2 Installing Intel MPSS This section outlines how to install, upgrade, and uninstall the software stack Intel MPSS installation The software stack is distributed in a single zip archive. 1. Download and extract the mpss-<version>-windows.zip archive. 2. Double-click the mpss-<version>.exe file. 3. Follow the on screen instructions to complete the installation Note: The default installation path is C:\Program Files\Intel\MPSS\. It can be changed during installation. Note: If a Windows* security prompt appears, allow the installation of the driver and software to proceed Unattended installation 1. Open the Command Prompt and navigate to the directory where the mpss- <version>-windows.zip file was extracted. For example: 20

21 Installation process overview User> cd C:\Users\<username>\Downloads\mpss-<version> 2. Enter the command below. User> "mpss-<version>.exe" /s /v /qn /V"/quiet /norestart" Unattended installation may take several minutes to complete Upgrade instructions Upgrading the Intel MPSS can be achieved by following instructions in Section Users may choose to manually uninstall the previous version or let the installer automatically search and remove previous release prior to installing the latest one. Note: Upgrading the software stack may restore the default configuration. Back up your configuration files and system images before performing the upgrade Installing Windows* cross-sdk The SDK for the coprocessor s native compiler is included in the installation zip file package. It also includes a squashfs repository image with supplementary software for the coprocessor. The SDK is required in order to compile and run applications for the coprocessor. Follow the steps below to install the SDK: 1. Extract the mpss-<version>-windows.zip file. 2. Install the software stack as in Section Double click the mpss-essentials-<version>.exe file. 4. Follow the on screen instructions to complete the installation. Installing the SDK is mandatory when using offload programming directives or cross compilers. It is unnecessary to uninstall previous software versions. The installer will automatically remove previous versions prior to installing the current one. Note: The Windows SDK does not contain header files necessary for cross-compiling Linux kernel netfliter modules Uninstalling Intel MPSS To uninstall the software stack open the Control Panel, choose Programs and features and remove the Intel(R) Xeon Phi(TM) coprocessor application. 3.3 Updating coprocessor s firmware It is required to update coprocessor s firmware after each installation or update of the software stack. Current firmware is distributed with the software stack installation. The readme-windows.pdf file (distributed within package mpss-<version>windows.zip) lists the versions of the included firmware. 21

22 Installation process overview Note: Running Intel MPSS with incorrect firmware version is not supported and may lead to erratic behavior. Follow the instructions below to update the coprocessor. Note, however, that these instructions will not work if the flash files (i.e. files with the.hddimg suffix) were moved or deleted from their installation directory. 1. Check the status of each coprocessor. Admin> micctrl -s If the status of every coprocessor is ready to boot, proceed to step 2; otherwise, reset the coprocessors. Admin> micctrl -rw 2. Verify the version of firmware installed on the coprocessor: Admin> micfw device-version If the versions are the same as versions described in Section of the readmewindows.pdf file, the next steps might be skipped. Note: When firmware installed is below to Intel MPSS version, step 3 has to be executed twice to complete the SMC firmware update. 3. Update the firmware of each coprocessor: Admin> micfw update all Or update only a specified coprocessor: Admin> micfw update micn Once the update process completes the state of coprocessors will be changed to ready to boot. Note: Do not execute any other applications or modify any coprocessor s state while micfw is executing. Note: micfw might fail to update coprocessor s firmware if it hadn t been consequently updated with each subsequent version of the software stack. 4. Perform the cold host system reboot to apply all changes. 3.4 Initial coprocessor boot Starting the coprocessor After successful completion of the previous steps, the coprocessor(s) can be booted. Enter the following command: Admin> micctrl --start 22

23 Installation process overview The call to micctrl exits when it determines the coprocessor either booted successfully or failed to boot. You can check the coprocessor s status to verify that booting was successful. Run the following command: Admin> micctrl s The command s output should indicate that coprocessor is online. Running the above commands also initiates the default per-coprocessor configuration. Note: The default configuration boots the coprocessor when the host driver loads, causing it to boot during the host OS startup Initial configuration Two files contain coprocessor configuration parameters: C:\Program Files\Intel\MPSS\default.conf contains configuration parameters parsed to every coprocessor in the system. C:\Program Files\Intel\MPSS\micN.conf contains configuration parameters for a specific coprocessor. The configuration files are created automatically when the micctrl command is run for the first time after the initial installation of the software stack. The micctrl command also creates configuration files if they were deleted from the host file system or after restoring the default configuration. You can modify the default configuration if it does not meet the requirements of your system (refer to Section 4 for more information). Reboot the coprocessor to apply the new configuration: Admin> micctrl --stop Admin> micctrl --start Specifying administrator s SSH key Access to the coprocessor is provided through the secure shell utilities. The initial login can only be authenticated with an SSH key pair. Section contains instructions on generating SSH keys and accessing the coprocessor through SSH using PuTTy tools. Specify the location of the administrator s public SSH key in the default.conf configuration file. The default value is: RootPublicKey %INTEL_MPSS_HOME%\filesystem\ssh\id_rsa %INTEL_MPSS_HOME% is an environment variable created by the software stack indicating its installation directory (the default installation path is C:\Program Files\Intel\MPSS). 23

24 Installation process overview The key will be copied to every coprocessor s file system during each boot. The administrator can then use their private key to log in to the coprocessor using the username root. 3.5 Validating Intel MPSS installation Intel MPSS provides utilities to perform basic tests to validate the installation. Section 4.4 provides troubleshooting advice if problems are encountered during software stack installation and usage. The micinfo tool provides information about the host and the coprocessor hardware and software. It can be used, for example, to verify that your coprocessor is running the correct firmware version. Note that certain information is only available when executing micinfo with root privileges: Admin> micinfo More information can be obtained by running the micinfo --help command. The miccheck tool performs sanity checks on systems containing coprocessors. Admin> miccheck Obtain more information on the utility by running the miccheck --help command. 24

25 Coprocessor OS configuration 4 Coprocessor OS configuration The coprocessor operating system requires a kernel and a file system image. These components are installed into the host s file system as part of the software stack. The kernel command line is constructed based on a set of configuration files on the host and is passed to the coprocessor s kernel during boot. Note: Configuration parameters and procedures for the previous generation Intel Xeon Phi coprocessor x100 are not supported in this version of the software stack. 4.1 Configuration files Configuration is controlled by parameters in files located in %INTEL_MPSS_HOME%, which by default is C:\Program Files\Intel\MPSS\. The default.conf file contains configuration parameters shared by all coprocessors in the system. Additionally, each coprocessor has an associated micn.conf configuration file, where N is the integer ID of that coprocessor (for example: mic0.conf, mic1.conf, etc.). Note: If the same parameter is present in default.conf and any micn.conf file, the latter takes precedence in configuring the coprocessor. Each configuration file contains a list of parameters and their arguments. Each parameter must be on a single line. Comments begin with the # character and terminate at the end of the line. There are several configuration parameter categories: 1. Parameters that configure the coprocessor Linux* kernel. 2. Parameters that configure the root file system. 3. Parameters that configure the boot command line. 4. Parameters that configure the host and coprocessor virtual Ethernet connection Updating configuration files Executing the micctrl --restoredefaultconfig command after upgrading the software stack updates all deprecated or changed configuration parameters. Note: Running the above command may overwrite your configuration files and file system images. 4.2 Boot configuration To boot a coprocessor, micctrl must: Determine the kernel to boot. Identify and/or build the file system image. Build the kernel command line. 25

26 Coprocessor OS configuration Parameters in the default.conf and micn.conf configuration files are evaluated for this purpose. The following sections describe parameters that are evaluated for this purpose Specifying the Linux* kernel The KernelImage and KernelSymbols parameters in micn.conf files determine the coprocessor OS boot image and its associated system address map file. Their default values are: KernelImage %INTEL_MPSS_HOME%\filesystem\bzImage.bin KernelSymbols %INTEL_MPSS_HOME%\filesystem\System.map Coprocessor-side kernel command line parameters Additional kernel command line parameters can be provided in the KernelExtraCommandLine configuration parameter in the default.conf configuration file. Refer to Appendix A.4.1 for more information. The complete list of kernel parameters is available at Coprocessor s root file system The RootFsImage parameter in micn.conf files specifies the coprocessor s root file system image. During initialization the software stack creates the %INTEL_MPSS_HOME%\filesystem\micN.ext4 file system image files for every coprocessor. Each of those files has a default size of 1GB Coprocessor s file system on diskless hosts Coprocessors installed in hosts with network storage or with limited storage capabilities can be configured to use a single root file system image. Multiple coprocessors can use a single file system image in the read-only mode and use it to create a file system in their RAM. Note that RAM file system will not persist across coprocessor reboots. To establish this configuration the RootFsImage parameter in the coprocessors micn.conf files should indicate the same file system image. Additionally, the mic_root_in_ram=1 option should be set in the KernelExtraCommandLine parameter in the default.conf (to apply to all coprocessors) or in the micn.conf file (to apply to a specified coprocessor) Automatic boot The AutoBoot parameter controls whether coprocessor is booted automatically during the host system startup. By default, this parameter is present in all micn.conf files, and set to Enabled. Set the parameter value to Disabled in any micn.conf file to change the default behavior and disable the automatic boot feature for that 26

27 Coprocessor OS configuration coprocessor. Remove the parameter from micn.conf files and add it to the default.conf file to enable or disable this feature for all coprocessors in the system. If the parameter is not present in any configuration file the behavior is as for AutoBoot Disabled Coprocessor shutdown The following commands can be used to reset, reboot or shut down the coprocessor: micctrl -r resets the coprocessor to the ready to boot state; micctrl -R reboots the coprocessor to the online (booted) state; micctrl -S - shuts down the coprocessor; 4.3 Basic administration tasks This section discusses how to perform some common administration tasks, such as adding and removing users and groups. Note that the initial configuration of the coprocessor OS only creates the root user account, and does not parse any user credentials to the coprocessor file system. This data can be provided manually in the same manner as in a standard Linux* distribution Adding groups and users to the coprocessor s file system Once logged in to the coprocessor, the administrator can use standard Linux* commands to add groups and users to its file system. [micn]# groupadd <group> [micn]# adduser <user> The <group> and <user> elements specify the group and username to be added. Invoke the commands with the --help option for more information Removing users and groups from the coprocessor s file system Removing a group or user from the coprocessor s file system is accomplished through the standard Linux* commands: groupdel and userdel. [micn]# groupdel <group> [micn]# userdel <user> The <group> and <user> elements specify the username and group to be deleted. Note that the home directory of the user will still exists in /home. Using the r option with the userdel command removes the user s home directory. It can also be removed manually with the command below. [micn]# rm rf /home/<user> 27

28 Coprocessor OS configuration Specifying the host SSH keys Users can run the command below to transfer their public SSH key to the coprocessor. This operation allows logging in to the device using a private key. Refer to Section for details on generating user SSH keys and using the PuTTY tool to establish ssh connection to coprocessors. Note: This operation requires the user to have an account in the coprocessor s file system. [micn]$ echo PUBLICKEY >> ~/.ssh/authorized_keys PUBLICKEY indicates the OpenSSH format public key. This command adds the specified public key to the authorized_keys file in the coprocessor file system. This command must be executed from a terminal prompt logged in as the desired user. 4.4 Troubleshooting This section describes using the Windows* Event Viewer, which may help troubleshooting the following events: The coprocessor times out during the boot process; The coprocessor does not start; The micctrl utility logs information related to events in the Windows* Event Viewer. To open the event viewer, click the Start button and type Event Viewer. In the left column expand Applications and Services and double-click MPSS Log. The following window will be displayed: This event log should be consulted in the event that any observable failures occur while starting or stopping coprocessors. Failures include timeouts and errors running micctrl --start, micctrl--stop, micctrl -r, and micctrl -b. 28

29 Coprocessor OS configuration The event log can be saved to a file from the command line with the command below. Admin> wevtutil qe /f:text "MPSS log" > output.txt It is also possible to print the software stack related information from Event Viewer on the console with the micctrl tool. Admin> micctrl -e 29

30 Network Configuration 5 Network Configuration The coprocessor does not have hardware Ethernet capabilities. Instead, virtual Ethernet drivers emulate Ethernet devices to enable a standard TCP/UDP IP stack on the host and coprocessor. Parameters in each coprocessor s micn.conf configuration file are consulted to create a virtual Ethernet interface for each coprocessor. Those parameters and their syntax are described in detail in Appendix A.5. During the software stack installation a static pair network configuration is established; see Section for a full description. 5.1 MAC address assignment The coprocessor has no pre-assigned MAC address because it lacks hardware network interface. Therefore, a MAC address must be generated and assigned to each virtual network device. During the coprocessor s boot, micctrl generates MAC addresses for the network endpoints. Those addresses are based on serial numbers of the devices. This behavior can be changed by specifying the HostMacAddress and CardMacAddress configuration parameters in the micn.conf files. Their syntax and supported values are described in Appendix A Host firewall configuration If the coprocessor s host system firewall is enabled, it may require configuration to allow NFS mounting of host exports. NFS requires access to the following ports: tcp/udp port RPC 4.0 portmapper tcp/udp port nfs server 5.3 Supported network configurations The supported network configuration schemes allow coprocessors to operate in a wide range of network environments. The static pair topology creates a private subnetwork between the host and each coprocessor (Section 5.3.1). The bridged topologies: o o Internal bridge allows coprocessors to communicate with the host and with each other (Section ). External bridge additionally allows coprocessors to communicate with external compute nodes (Section ). 30

31 Network Configuration Static pair Static Pair Configuration creates separate private network between the host and each coprocessor with IP addresses assigned to every network endpoint. Static pair is the default network configuration. It is sufficient for Intel C++ and FORTRAN compiler pragma-based offload computation on a standalone workstation. The static pair configuration creates a separate private network between the host and each coprocessor, and assigns an IP address to each of the network endpoints. Figure 4 depicts this configuration. A private network is configured between the host and each coprocessor. Notice that mic0 and mic1 are on separate subnets. Figure 4 Static pair configuration Static pair configuration To establish or modify this configuration edit the micn.conf file. The values of the HostNetworkConfig and CardNetworkConfig parameters should resemble the examples below: HostNetworkConfig inet static address= netmask= CardNetworkConfig inet static address= netmask= gateway= The IP addresses of the host and each coprocessor must be on the same subnet. A descriptor of each coprocessor endpoint should be added to the host s %WINDIR%\System32\drivers\etc\hosts file to associate IP addresses with each coprocessor s hostname. For example: mic mic1 31

32 Network Configuration Bridging Similarly a descriptor of the corresponding host endpoint should be added to each coprocessor s /etc/hosts file to associate the host s endpoint IP address with the host s hostnames. For example, mic0 s /etc/hosts might contain: localhost.localdomain localhost ::1 localhost.localdomain localhost host mic0 Note that in this example /etc/hosts includes descriptors of both the host endpoint and the local endpoint. Endpoints are brought up during the coprocessor s boot. At this point the ipconfig command output executed on the host should be similar to: Ethernet adapter mic0: Connection-specific DNS Suffix. : Link-local IPv6 Address..... : fe80::b57a:a0e0:c5eb:5b24%29 IPv4 Address : Subnet Mask : Default Gateway : A network bridge is a way to connect two Ethernet segments or collision domains in a protocol independent way. It is a Link Layer device which forwards traffic between networks based on MAC addresses and is therefore also referred to as a Layer 2 device. The software stack directly supports two types of bridged networks described below Internal bridging An internal bridge allows the connection of one or more coprocessors within a single host system as a subnetwork. In this configuration, each coprocessor can communicate with the host and other coprocessors in the platform. Figure 5 shows an example internal bridge configuration. 32

33 Network Configuration Figure 5 Internal bridge configuration Such network configuration could, for example, be used to support communication between the ranks of an MPI application that is distributed across coprocessors and host. The internal bridge requires setting a new host bridge endpoint and connecting coprocessor network adapters to it Internal bridge configuration Open Network and Sharing Center and select Change adapter settings. A window will show displaying available network adapters that should contain virtual network adapters for each coprocessor in the system (named mic0, mic1, etc.). Select adapters you wish to connect to the bridge and choose Bridge Connections from the context menu. A network bridge will be created automatically with default configuration and chosen adapters will be connected to it. Right click the Network Bridge, which should be now visible in the window, and select Properties. The new window displays a list of network adapters connected to the bridge, allows to remove or add additional adapters and contains configuration options for the bridge. 33

34 Network Configuration Select Internet Protocol Version 4 (TCP/IPv4) and click Properties. Mark the Use the following IP address option and provide IP address and subnet mask for the network bridge (e.g , ). Confirm your changes by clicking the OK button. Set correct IP address for each coprocessor connected to the bridge by modifying the CardNetworkConfig parameter in their respective configuration files, netmask value should be the same for all coprocessors. The examples below show configuration for coprocessors mic0 and mic1. Coprocessors will use their new configuration after a reboot. 34

35 Network Configuration mic0.conf: CardNetworkConfig inet static address= netmask= gateway= mic1.conf: CardNetworkConfig inet static address= netmask= gateway= Connecting coprocessors to the bridge causes the HostNetworkConfig parameter to be ignored. Executing the ipconfig command in the command prompt will display details on the network configuration on the host. Once logged in on the coprocessor, use the ifconfig command to view information about network configuration. To change the MAC address of the network bridge open its properties, click Configure, choose the Advanced tab, select the Network Address property and provide the MAC address. By default, the network bridge obtains its MAC address from one of the connected adapters External bridging The external bridge configuration bridges one or more coprocessors to an external network. This is a typical configuration required when coprocessors are deployed in a cluster to support remote communication between them and/or CPUs across different compute nodes. Figure 6 depicts a cluster in which the coprocessors on each host node are bridged to an external network. The IP addresses in such a configuration can be assigned statically or by a DHCP server on the network, but must generally be on the same subnet. In most cluster environments it is usually recommended to employ static IP schemes, which allows to correlate nodes with their known IP addresses. InfiniBand* based networking is not shown in this figure. InfiniBand* based networking will usually provide significantly higher bandwidth than the IP networking supported by the Intel MPSS Virtual Ethernet driver. Many clusters use Ethernet* networking for low bandwidth communication such as command and control, and use InfiniBand* networking for high bandwidth communication as application data transfer. 35

36 Network Configuration Figure 6 External bridge configuration External bridge configuration Note: Make sure that you have provided a large enough IP address pool to accommodate nodes of the externally bridged networks. Add coprocessors virtual network adapters to a network bridge and connect the bridge to the external network. a) Static IP address assignment Follow the instructions in Section Remember to reboot the coprocessors if during configuration you modified their configuration files. b) IP address assignment by DHCP server 36

37 Network Configuration Open the properties window of the network bridge and in the Internet Protocol Version 4 (TCP/IPv4) properties select Obtain an IP address automatically and Obtain DNS server address automatically. To allow coprocessors to obtain their IP addresses from the DHCP server modify the CardNetworkConfig parameter in their configuration files. The parameter should have the following value: CardNetworkConfig inet dhcp Reboot the coprocessors to apply the new configuration. 5.4 Interacting with the coprocessor The following sections describe how to generate SSH keys, access the coprocessor and transfer files to its file system. The examples are based on the PuTTYgen key generation tool, PuTTY SSH client and winscp SCP client External tools Download the latest version of PuTTY and PuTTYgen from WinSCP can be downloaded from Install the tools in the same location as the external tools: C:\Program Files\Intel\MPSS\bin SSH access to the coprocessor Steps in this section show how the system administrator can establish a key-based SSH access to the coprocessor. To allow non-root users to log in to the coprocessor create additional user accounts in its file system and specify each user s public key in their ~/.ssh/authorized_keys file. Section explains how to perform these tasks. Note: By default, only key-based authentication can be used to log in to the coprocessor. Any attempt to establish SSH connection using password will be rejected by the ssh daemon on the coprocessor. This behavior can be changed by modifying the /etc/ssh/sshd_config file. 1. Open PuTTYgen (puttygen.exe). In the program window click Generate. After completing the on-screen instructions the following screen is displayed. 37

38 Network Configuration 2. Click Save private key and save the private key in a secure location not readable by other users. This key will be used to connect to the coprocessor. Optionally, save the key with a passphrase, or click Yes to ignore the warning. 3. Select Save public key to save the public key in a file. 4. Set the value of the RootPublicKey parameter in the default.conf configuration file to the location of your public key. The key will be propagated to each coprocessor during boot. 5. Open PuTTY.exe. In the Category box, expand Connection -> SSH and select Auth. In the Private key file for authentication field, browse to the private key saved in step 2. 38

39 Network Configuration 6. In the Category box, select Session. In the Host Name (or IP address) field, enter the coprocessor s IP address (mic0 shown). You may save the configuration under custom name by providing the name in Saved Sessions box and clicking Save. This allows for easy reconnection to the coprocessor without specifying the private key each time. Click Open, this will establish connection with the coprocessor mic0. 7. When the session opens to the coprocessor enter the username root. You should successfully log in without the need to provide password. 39

40 Network Configuration After completing these steps, a program such as WinSCP can be used to transfer files to and from the coprocessor Transferring files to and from the coprocessor This section describes using the WinSCP tool (version shown) to transfer files to and from the coprocessor. Instructions here assume that key-based authentication is established. 1. Launch WinSCP and select File Protocol: SCP. To transfer files to coprocessor mic0 type in the Host Name field and your username as specified in the coprocessor s file system in the User Name field (the administrator should enter root). Do not enter a password. Click Advanced to display the Advanced Site Settings dialog. 2. In the Advanced Site Settings dialog, select SSH -> Authentication. In the Private key file field, enter the full path to the private key file generated above (or click [ ] to browse to the file). 40

41 Network Configuration 3. Select Environment -> SCP/Shell. In the Shell field, select /bin/bash. Click OK to accept new configuration and return to the previous window. 41

42 Network Configuration 4. Optionally, click Save to save the new site to the tree in the left pane. Click Login to use the tool. 42

43 Managing the coprocessor s file system 6 Managing the coprocessor s file system The persistent files system (PFS) is the root file system of the Intel Xeon Phi coprocessor x200 is maintained on the host with a virtual block device on the coprocessor. This device redirects read/write requests to and from the host over PCIe. This solution enables the user to manage the coprocessor s file system much like the one in a standard Linux* OS. The mechanism is depicted in Figure 7 Figure 7 Persistent file system concept 6.1 Adding persistent files to the coprocessor s file system Files added to the coprocessor s file system during normal use are stored persistently by default. A user has several options to make changes to the persistent file system. Supplementary software for the coprocessor is installed on the host with the Windows* cross-sdk (mpss-essentials-<version>.exe). 43

44 Managing the coprocessor s file system Using the smart package manager The smart package manager allows users to install additional software on the coprocessor. The squashfs repository image on the host (e.g. the repository installed with the Windows* cross-sdk) is passed to the RepoFsImage configuration parameter in each micn.conf file. The repository image is automatically detected and configured by the smart package manager. Available packages can be installed by running the smart install <package> command, for example: [micn]# smart install perf You can find more information on the smart package manager at Refer to Appendix A.3.2 for more information on the RepoFsImage parameter Editing files interactively Editing files interactively assumes that the user logs in to the coprocessor using SSH, and introduces desired changes directly in the coprocessor s OS in the same manner as in a standard Linux* distribution Transferring files over SSH Copying a file over SSH assumes that the user is able to log in to the coprocessor using SSH. Copy a file from the host to the coprocessor with WinSCP, pscp, or another similar utility. Refer to Section for instructions on transferring files using WinSCP Installing RPM packages (root only) Installing an RPM package assumes that the user has access to the root account in the coprocessor s file system and is able to log in using SSH. To install a package, copy the RPM file(s) to the coprocessor s file system (over SSH/SCP) and issue the following command: [micn]# rpm ihv <package_name> 6.2 Virtual block devices Users can specify additional block devices with supplementary data or software for the coprocessor. The BlockDevice configuration parameter allows to specify up to 6 additional block devices for each coprocessor. 44

45 Managing the coprocessor s file system Each virtual block device must be backed by a supported 1 image file on the host and cannot be shared by multiple coprocessors. Each specified block device will be identified by the coprocessor as a /dev/vd[a-z] device. Example below shows how to specify a read/write block device named storage1 backed by a file at <image file location>. BlockDevice name=storage1 mode=rw path=<image file location> Once the block device is specified use the commands below create a mount point and attach the device to the coprocessor s file system. [micn]$ mkdir p <mount point> [micn]$ mount /dev/vd[a-z] <mount point> The software stack creates symbolic links to each device at /dev/mpss_mapper<block name>, allowing users to refer to block devices by their designated names. The commands below show how to mount a block device named storage1. [micn]$ mkdir p <mount point> [micn]$ mount /dev/mpss_mapper/storage1 <mount point> Add the virtual block device to the /etc/fstab file in the coprocessor s file system to mount it automatically upon every boot. Refer to Appendix A.3.3 for more information on the BlockDevice parameter. 6.3 Host file system share This section shows how to set up, configure and mount an NFS Share on Windows* Server 2012 R2 and Windows* Server Section through show how to use the Server Manager user interface to set up an NFS share. For an alternative method using PowerShell* cmdlets, skip to Section Note: Feature described in this section is not available on Windows* 8.1 and Windows* 10 1 Supported image files include, among others: cpio, ext2, ext3, ext4, hddimg, iso, squashfs, tar, vdi and vmdk 45

46 Managing the coprocessor s file system Using the server manager Adding the NFS server 1. Start Server Manager. On the dashboard, in the Manage menu select Add Roles and Features. 2. In the Add Roles and Features Wizard, click Installation type in the left column. Select Role-based or feature-based installation. 46

47 Managing the coprocessor s file system 3. Click Server Selection and Select a server from the server pool. Select the server. 4. Click Server Roles. Under Roles, expand File and Storage Services, then expand File and iscsi Services. Select Server for NFS. 47

48 Managing the coprocessor s file system 5. A confirmation dialog box will appear. Click the Add Features button. 6. Returning to the Add Roles and Features Wizard, select Confirmation in the left column. Click the Install button. 48

49 Managing the coprocessor s file system 7. In the Results pane, confirm that the installation was successful Provisioning a folder for the NFS share 1. In Server Manager, click File and Storage Services in the left column of the Dashboard. 2. Select the desired server from the list. Click Shares in the left column. 49

50 Managing the coprocessor s file system 3. Click To create a file share, start the New Share Wizard. 4. In the New Share Wizard, click Select Profile and select NFS Share - Quick. 5. Click Share Location and select the desired server from the list. Specify a path for the share. 50

51 Managing the coprocessor s file system 6. Click Share Name and specify a name for the share (click OK if prompted to create the share location directory). 7. Click Authentication. Select No server authentication, Enable unmapped user access, select Allow unmapped user access by UID/GID. 51

52 Managing the coprocessor s file system Specifying the NFS share permissions 1. Click Share Permissions and click Add. In the Add Permissions dialog box, change the Share Permissions field for the All Machines group to No Access. This prevents unspecified hosts from accessing the NFS share. Click Add. 2. Return to the New Share Wizard and click Add. Now, select Host in the Add Permissions dialog box. Specify a host and set its Share Permissions field to Read/Write. Select Allow root access. Click Add. 52

53 Managing the coprocessor s file system 3. Return to the New Share Wizard. Review the Share Permissions settings. Verify the Host has Read/Write permissions and Root Access is Allowed. 4. In the left column, click Permissions then click Confirmation. Confirm your settings and click Create. 53

54 Managing the coprocessor s file system 5. Confirm your results and close New Share Wizard and Server Manager. 6. Proceed to Section Using PowerShell* cmdlets to create an NFS share (optional) The tasks in previous sections can be accomplished using the built-in PowerShell* NFS cmdlets in Windows* Server 2012 R2 and Windows* Server 2016, as shown below: Note: The backslash \ character at the end of a line indicates continuation of a command to the next line. It is not part of the command itself. 1. Add the NFS server: User> Add-WindowsFeature FS-NFS-Service 2. Provision the directory to be shared: User> New-Item C:\MY_NFS_TEST -type directory User> New-NfsShare -Name nfs-tests \ -Path C:\MY_NFS_TEST -EnableUnmappedAccess $True \ -Permission no-access -Authentication sys 3. Grant read/write and root access permissions for Host : User> Grant-NfsSharePermission Name nfs-tests \ -Permission readwrite -ClientName \ -ClientType Host -AllowRootAccess $True 4. Confirm share permission settings: User> Get-NfsSharePermission Name nfs-tests 5. Proceed to Section

55 Managing the coprocessor s file system Mounting the NFS share Note: Refer to Section for information on downloading and installing PuTTY tools. 1. Change to the directory where PuTTY tools are installed. User> cd <PATH_TO_PUTTY_TOOLS> 2. Log in to the coprocessor as root; example below shows logging in to coprocessor mic0. User> putty.exe -ssh -i <PATH_TO_PRIVATE_KEY> root@ <PATH_TO_PRIVATE_KEY> indicates the location and name of the private key file associated with this coprocessor. Note: If prompted for a password, see Section for instructions on setting up key-based SSH authentication 3. While logged in to the coprocessor, add a directory with the same name as the NFS share: [micn]# mkdir /tmp/nfs-tests 4. Mount the NFS share in the coprocessor s file system. The nolock option is required in this step: [micn]# mount -t nfs -o nolock \ :/nfs-tests /tmp/nfs-tests 5. Navigate to the directory that you created on the coprocessor. Verify the contents of the NFS share are visible. [micn]# cd /tmp/nfs-tests 6.4 CIFS share on Windows* This section provides instructions on how to set up and provision a CIFS (Common Internet File System) share between the Windows* host and coprocessor Configuring a CIFS share To mount a CIFS share, you must grant share permissions to Everyone and turn off password protected sharing. 1. Open the Network and Sharing Center, click Change advanced sharing settings, expand All Networks and enable the Turn off password protected sharing option. 2. Choose a folder you wish to share and grant appropriate permissions for that folder to Everyone. 55

56 Managing the coprocessor s file system a. Open properties of the folder, select the Security tab and click Edit to change the folder permissions. b. Click Add, type Everyone in the Enter the object names to select field and click Ok. 56

57 Managing the coprocessor s file system c. Grant appropriate permissions to the Everyone group, which should now be visible in the Groups or user names filed. 3. Choose the Sharing tab in the properties window of the folder you wish to share and click Share. Add Everyone to the share group, assign appropriate permissions and select Share 4. Click Advanced Sharing in the Sharing tab and select Share this folder. 57

58 Managing the coprocessor s file system 5. Navigate to Permissions and assign appropriate permissions to Everyone. Note: You have to enable sharing permissions in both Share and Advanced Sharing. The current mount utility on the coprocessor does not support nested host share paths Mounting a CIFS share The root user can manually mount and umount password protected shares on the coprocessor. To mount a CIFS share, issue the following command on the coprocessor. [micn]# echo <SHAREPATH> <MOUNTPOINT> cifs \ guest,file_mode=0777,dir_mode= >> /etc/fstab Use the command below to mount a CIFS share for a specified user or group, for example Everyone. [micn]# echo <SHAREPATH> <MOUNTPOINT> cifs username=everyone,\ password=,file_mode=0777,dir_mode= >> /etc/fstab <SHAREPATH> is the network addressable CIFS share. As an example, assuming that mic0 s host network adapter has the static IP address , the share path would be // /<path to share>. The <MOUNTPOINT> is the directory in the coprocessor s file system to mount the CIFS share to. For example, the mount point could be /mount/host_share. The directory used for the mount point must exist before mounting the CIFS share. This command adds an entry to the coprocessor s fstab. 58