发布时间 : 星期一 文章Windows CCS and Linux dual boo.doc更新完毕开始阅读dbfdf1eab8f67c1cfad6b81c
There are other configurations and deployment methods available for a dual-boot HPC cluster, but these were not verified for this document. Manual installation is not described here, but it can be easily derived from the provided scripts and installation guidelines. The installation methods and configurations that you choose will depend on the number of compute nodes that you use, and on the order in which you choose to install the operating systems. Appendix 2 provides information about installation options including Windows Deployment Services (WDS); Appendix 3 gives an example of Windows Compute Cluster Server 2003 deployment with the Automated Deployment Services (ADS).
In this document, it is assumed that all compute nodes and the required infrastructure servers of the compute cluster do not operate with different OS versions concurrently. This is technically feasible, but it requires a sophisticated infrastructure setup that is beyond the
scope of this document. This document does not describe how to dual boot the cluster‘s head node or the infrastructure servers. For these servers, it is possible to use a boot menu to choose the desired OS at startup. Infrastructure servers may require additional partitions in order to run the two operating systems.
The focus of this document is the deployment of the Windows Compute Cluster Server 2003 OS and SUSE Linux Enterprise Server 10 SP1 to the compute nodes of a cluster and the operation of the cluster in a dual-boot environment. Configuration of Linux-based and
Windows-based HPC clusters is not described. For information on the installation of an HPC cluster, see the related links in Appendix 5.
This document is intended for Linux administrators and Windows IT professionals who are familiar with HPC and who have some familiarity with Windows Compute Cluster Server 2003. For more information about the products and methods described, see the links in Appendix 5.
Dual Boot: Windows Compute Cluster Server 2003 and SUSE Linux Enterprise Server 10 5
Prerequisites
Before you begin the dual-boot configuration and deployment, be sure to read through all of the procedures and verify that your configuration meets all requirements. You should also become familiar with the product documentation for the operating systems and for the deployment methods (see Appendix 5 for related links). Be sure to test the deployment procedures on one node before deploying the operating systems to the entire cluster. You should also verify the operation of your cluster by running a small cluster-aware application.
Note: This document describes one particular installation, in which all infrastructure services,
?
such as DNS, Active Directory, and DHCP, were installed on the compute cluster head node. Your configuration may be different, and these services may be installed on a dedicated infrastructure server. You may therefore need to modify the procedures described.
Mixed Compute Cluster Configuration
A compute cluster uses several types of servers, some required and some optional. The servers that you need will depend on your cluster configuration.
In this installation, a dedicated deployment server was used for the SUSE Linux Enterprise Server deployment. RIS, as available on the WCCS head node by default, was used to deploy the Windows Compute Cluster Server 2003 OS. An infrastructure server was not used; all services were placed on the head node.
Note: If you are not building a cluster for test or evaluation purposes, you may have services such as DNS, Active Directory, and DHCP, on dedicated servers, such as the infrastructure server described below.
Compute Nodes
? ? ?
Windows Compute Cluster Head Node
One server is required to act as the compute cluster head node. This server can be configured to act as a compute node, as well.
?
The head node can provide the required infrastructure services, such as DNS, Active Directory, and DHCP; this is an option in a test or evaluation environment, or in a small cluster running on an isolated network.
Note: Windows Internet Connection Sharing (ICS) can assign IP addresses to the nodes, but you can use a DHCP server on the head node or on the infrastructure server instead. ICS has an address range limit of 192.168.0, and limits the number of nodes in a cluster to 250.
The compute nodes‘ BIOS must be configured so that the Preboot Execution Environment (PXE) is the first boot option.
If the compute nodes have two or more Network Interface Cards (NICs), PXE must boot only on the interfaces connected to the compute cluster‘s private network. Manufacturers‘ network drivers may be required for the RIS installation.
Dual Boot: Windows Compute Cluster Server 2003 and SUSE Linux Enterprise Server 10 6
? ?
For RIS, the head node must have two NTFS partitions on the hard disk and must be configured with at least two NICs.
Configuration of the infrastructure of the head node is not described in this document. Refer to the Windows WCCS documentation for information on infrastructure configuration; for the link, see Appendix 5.
Infrastructure Server
This server is optional in a test or evaluation environment.
?
The infrastructure server can provide DNS, DHCP, file services and Active Directory services if these services are not provided by the head node due to configuration and performance concerns.
The infrastructure server can be configured with two NICs, one for the private network and one for the public network.
?
Software Requirements and Versions Used
Windows Compute Cluster Server
This installation used Windows Compute Cluster Server 2003 with Service Pack 1 (SP1), which consists of Windows Server 2003 Compute Cluster Edition (CCE) with Service Pack 2 (SP2) and the Microsoft Compute Cluster Pack (CCP) with Service Pack 1 (SP1). The head node used Windows Server 2003 Standard x64 Edition with Service Pack 2(SP2).
The CCP is a combination of interfaces, utilities, and management infrastructure; it can be purchased separately and deployed to an existing server running Windows Server. Other
supported operating systems for both head node and compute nodes include Windows Server 2003, Standard x64 Edition; Windows Server 2003, Enterprise x64 Edition; Windows Server 2003 R2 Standard x64 Edition; and Windows Server 2003 R2 Enterprise x64 Edition.
SUSE Linux Enterprise Server
For Linux installation and deployment, SUSE Linux Enterprise Server 10 SP1 was used and verified on x64 hardware with serial attached storage (SAS) drives and included the MPICH (MPI) RPMs. The GRand Unified Bootloader (GRUB) was also used.
Hardware Used
This installation used the Microsoft Enterprise Engineering Center‘s 64-node Skagit cluster for the verification of this document. There are many possible hardware configurations, and you may need to modify the procedures based on the equipment that you use.
The Skagit cluster uses:
?
4 HP BladeSystem c7000 enclosures, each with:
o At least one GbE2c Ethernet Blade Switch
o At least one 4X DDR IB Switch Module for HP c-Class BladeSystem (IFB)
?
64 HP ProLiant BL465c G1 Server Blade nodes, each with:
o 2 dual-core AMD Opteron Processors 2216 HE at 2.40 GHz (1 MB L2)
Dual Boot: Windows Compute Cluster Server 2003 and SUSE Linux Enterprise Server 10 7
o 8 GB PC2-5300 DDR2 Registered DIMMs o 1 HP Smart Array E200i
o 2 73.4GB SAS Drives initialized with RAID
2 embedded NC370i Multifunction Gigabit Server Adapters
Prepare for Windows Compute Cluster Server 2003 and SUSE Linux Enterprise Server 10 SP1 Dual Boot
Before you begin the deployment and configuration of the compute cluster, verify the following: Check when completed Prerequisites The PXE boot option is disabled on all of the network adapters on the head node. If using a Windows infrastructure server, the PXE boot option on all NICs on the infrastructure server must be disabled. The PXE boot option must be disabled on the Linux deployment server. All drivers that are not integrated into the Windows setup files by default must be available. All of the compute cluster hardware must be set up and working. A naming schema must be provided. An IP address schema must be provided. The DHCP service must be installed on the compute cluster head node or on the infrastructure server. The DHCP scopes for the cluster‘s private and MPI networks must be created. The DHCP scope options (at a minimum, DNS servers and DNS domain name) must be created. The DHCP server must be authorized in Active Directory. The MAC addresses of the network adapters connected to the private network on all nodes must be available. These are required for configuring the Linux deployment server and the Windows DHCP service. The DHCP service must be configured to use static IP addresses for node identification. DNS must be installed and configured on the head node or on the infrastructure server. Active Directory must be installed on the head node or on the infrastructure server. The DHCP service must be configured so that it serves the nodes on both the private network and the MPI network. All required hotfixes must be available. Dual Boot: Windows Compute Cluster Server 2003 and SUSE Linux Enterprise Server 10 8