CTC: Getting Started-New User Guide

Welcome to the Cornell Theory Center! This guide covers what you'll need to know to get started using CTC's Velocity Cluster, and includes pointers to more information on each topic.

Table of Contents

Logging In
- Obtaining an Account
- Connecting to Velocity
- Velocity Configuration
- Downtimes
- Billing
- Passwords
- Mail
Working with Files
- The File Server
- Your Home Directory
- Protecting Your Files
- Tokens and "Permission Denied" Messages
- Backups
- Scratch Disk Space
- The MCP Utility
Running Programs
- Paths
- Applications Software, Compilers and Programming Tools
- Parallel Programming
- Running in Batch
Finding More Information
- World Wide Web Pages
- Status
- Help
- Consulting
- Workshops

1. Logging In
Obtaining an Account: To apply for time on Velocity, submit a proposal as described in the Allocations documentation. If your proposal is accepted, you will be given a login name for each member of your group and an allocation of wall-clock hours. For details see Allocations Policy.

Connecting to Velocity: Once you have a login name, you can connect to the login machines on Velocity (winlogin1 or winlogin2) with a Remote Desktop Connection or ssh. For details see "Accessing CTC Machines". Remote Desktop Connection will provide a Windows environment in a separate window and is the preferred mode for connecting to the login machines. If you want a UNIX-like environment, bring up a command prompt (under Programs/accessories). A quasi-Unix environment is available under Cygwin. Details are provided that will allow you to make Cygwin automatically available when you ssh to the login machines.

Velocity Configuration: The Cornell Theory Center maintains a number of different clusters of machines. For details see CTC Computing Resources.

Downtimes: Downtimes are taken Wednesday mornings on an as-needed basis. They are posted in advance on the Cluster Status page.

Billing: The billing rate is shown in the rate schedule. Use the vbalance command to check your account's balance. Time lost because of system problems beyond your control will be refunded, see the Refund Policy.

Passwords: At your first login you will be prompted to change your password. This is required for security purposes. CTC's password policy gives guidelines for selecting passwords that will be hard for someone else to guess. It also indicates password-related commands that must be issued prior to using the Velocity Scheduler (vsched) batch system and running parallel jobs using MPI.

Mail: There is no email service on the Windows cluster.

2. Working with Files

The File Server: Files that you create or store on Velocity are accessible from a Windows file server, logically named \\tc.cornell.edu\tc. These files on are accessible from any CTC machine.

Your Home Directory:
Your home directory is located at file://tc.cornell.edu/tc/Users/your_userid, where your_userid is your login name. This is mapped to the H: drive. At login, your home directory will appear as H:\Users\your_userid.

Protecting Your Files: The file servers use NTFS for their filesystem, which means that directory and file level permissions are available. To set or view permissions on a directory or file, you must either be the owner or have the Full Control permission. Using NTFS permissions, users can grant access to their home directory by giving other users any (or none) of the variety of permission levels that Windows offers. By default, your home directory can be accessed by you alone. Keep your directory structure in mind when assigning permissions. Other users must have at least Read permissions on all parent directories to the directory you are attempting to share.

If you are using Windows, set permissions by a right-click on a file or folder, then choose Properties | Security. If you are in DOS, help cacls gives information on modifying access control lists of files. This command works in a counter-intuitive manner and is not recommended. It is extremely easy to wipe out all permissions, so use it with caution.

Warning: Never give Full Control to anyone for your home directory, or any subdirectory; grant access privileges only as needed.

Full details on protecting your files and directories through permissions is here.

Tokens and "Permission Denied" Messages: To write a file or to read a protected file in Windows, you must have an authentication "token" issued by Windows's security system. A token is normally issued to you when you log in. This token identifies you as the username you used to connect to the file server, and therefore you inherit all permissions of that username. Note: This is not a kerberos token.

Backups: Every night all home directories are backed up.

Scratch Disk Space: Each of Velocity's batch machines has at least 18 Gigabytes of scratch disk space mapped to its t:\ directory. For details see CTC Computing Resources. This space is locally attached to each machine for faster access. You may use this temporary scratch disk space as work space for your jobs while they are running. When your job ends, copy all the data that you wish to keep from t:\ to some other location, such as your home directory.

Scratch space warnings: Files in scratch directories are not backed up. Currently files in t:\ on batch machines are not erased; you should explicitly delete sensitive files when your job ends. It is good housekeeping practice to delete all files from your t:\directory before releasing the machine(s) assigned to your job.

The MCP Utility: The mcp utility in H:\CTC Tools is used to copy a file to local disk on every machine that your parallel job is using. For more information on the mcp command, see the mcp page.

3. Running Programs

Paths: There are two paths in Windows: a system path and a user path. The system path is shared by everyone who logs into that computer and can only be modified by someone with administrator permissions. The user path is unique to each user account that logs into that computer, and can be modified by the account owner. The system path and the user path are concatenated to form the search path. The system path comes first, followed by the individual path and it is searched in that order. The search of the path stops as soon as the command is found. However you choose to set your path, remember that the names are case sensitive.

In Windows, set the user path by left-clicking the system icon (in the control panel) and choosing the "Advanced" tab, then click on "Environment Variables" to display both the system variables and the user environment variables, including paths. You should be able to create a user path or select and edit an existing user path. Changes made to the path are in effect immediately, but currently running applications will not know about the changes.

If you prefer to use command line prompts, type set path=%path%; (new paths to be appended). This will effect the current session only, which means that once this session is closed, the new path setting is gone. If there are two Command Prompts open, setting new path in one prompt will not effect the other. If you ssh to one of the login machines, and path settings are set in .bashrc files, the settings effect all the login ssh sessions under your account.

Applications Software, Compilers and Programming Tools: A complete list of the software available on the login machines, various types of compute machines and the Collaboratory is updated as new packages are added.

Programs can be compiled on your Windows machine, in the Collaboratory in Rhodes Hall, or on one of the login machines reached via a Remote Desktop Connection or ssh. For more information on writing programs, see Compiling and Porting Applications.

Parallel Programming: There are a variety of ways to do parallel programming on Velocity:

Native multithreaded applications - SMP
OpenMP can use SMP nodes
MPIPro is a way to use multiple nodes via message passing
MPIPro and Open MP together can combine smp and message passing
MPIPro can be used to run multiple processes on an SMP node and across multiple nodes. It automatically does message passing through shared memory on an SMP node and over the network between nodes.
Parallel tools are available such as MPIPro from MPI Software Technologies and OpenMP from Intel.

Running in Batch: Velocity Scheduler (vsched) is a job scheduler that distributes jobs to a cluster of workstations and/or to processors of a multi-processor machine. The scheduler decides when and how a batch job is run based on preferences set up by the user and system administrator(s). You can communicate with vsched using a few simple commands.

Jobs are submitted to the batch system via an .xml file. When a job is submitted, it is held in the queue until the scheduler can allocate the required resources to run the job. If the job is a batch job, then the script is executed immediately after resources have been allocated. Once the job has completed, the user will (optionally) be notified. If the job is an interactive job, then the user must log in to the machine(s) when the job has started, and execute his programs. You cannot specify which machines to run on, but you may specify your requirements; vsched will then choose appropriate machines.

Additional vsched documentation is available, including sample batch files and utilities.

4. Finding More Information

WWW pages: CTC's collection of web pages contains extensive documentation for all aspects of the system. If the menus do not lead you to the topic you are seeking, try the Search button at the top of the page.

Status: System status, including timely information such as system availability, schedule changes, and status of upgrades is available at
http://www.tc.cornell.edu/Services/Support/general_topics.cluster_status.info.htm

Help:

DOS: For a list of DOS commands, type Help at the command line. For help on any one command, type help command_name.
Windows: For help in Windows, choose Help from the Start menu.
Menu bars: Most menu bars include a Help button.
Cygwin: http://cygwin.com/faq/faq_toc.html

Consulting: If you have questions or encounter difficulties while using any CTC system, CTC consultants may be able to help. Please send electronic mail to consult or call (607) 254-8686 and press 2 to speak to a consultant.

Workshops offer intensive, hands-on instruction in the use of Theory Center resources. A diverse schedule of workshops offered in person at the Theory Center is supplemented by Virtual Workshops offered on the Internet.


Home Services Support

Getting Started-New User Guide

Welcome to the Cornell Theory Center! This guide covers what you'll need to know to get started using CTC's Velocity Cluster, and includes pointers to more information on each topic. Table of Contents Logging In Obtaining an Account Connecting to Velocity Velocity Configuration Downtimes Billing Passwords Mail Working with Files The File Server Your Home Directory Protecting Your Files Tokens and "Permission Denied" Messages Backups Scratch Disk Space The MCP Utility Running Programs Paths Applications Software, Compilers and Programming Tools Parallel Programming Running in Batch Finding More Information World Wide Web Pages Status Help Consulting Workshops 1. Logging In Obtaining an Account: To apply for time on Velocity, submit a proposal as described in the Allocations documentation. If your proposal is accepted, you will be given a login name for each member of your group and an allocation of wall-clock hours. For details see Allocations Policy. Connecting to Velocity: Once you have a login name, you can connect to the login machines on Velocity (winlogin1 or winlogin2) with a Remote Desktop Connection or ssh. For details see "Accessing CTC Machines". Remote Desktop Connection will provide a Windows environment in a separate window and is the preferred mode for connecting to the login machines. If you want a UNIX-like environment, bring up a command prompt (under Programs/accessories). A quasi-Unix environment is available under Cygwin. Details are provided that will allow you to make Cygwin automatically available when you ssh to the login machines. Velocity Configuration: The Cornell Theory Center maintains a number of different clusters of machines. For details see CTC Computing Resources. Downtimes: Downtimes are taken Wednesday mornings on an as-needed basis. They are posted in advance on the Cluster Status page. Billing: The billing rate is shown in the rate schedule. Use the vbalance command to check your account's balance. Time lost because of system problems beyond your control will be refunded, see the Refund Policy. Passwords: At your first login you will be prompted to change your password. This is required for security purposes. CTC's password policy gives guidelines for selecting passwords that will be hard for someone else to guess. It also indicates password-related commands that must be issued prior to using the Velocity Scheduler (vsched) batch system and running parallel jobs using MPI. Mail: There is no email service on the Windows cluster. 2. Working with Files The File Server: Files that you create or store on Velocity are accessible from a Windows file server, logically named \\tc.cornell.edu\tc. These files on are accessible from any CTC machine. Your Home Directory: Your home directory is located at file://tc.cornell.edu/tc/Users/your_userid, where your_userid is your login name. This is mapped to the H: drive. At login, your home directory will appear as H:\Users\your_userid. Protecting Your Files: The file servers use NTFS for their filesystem, which means that directory and file level permissions are available. To set or view permissions on a directory or file, you must either be the owner or have the Full Control permission. Using NTFS permissions, users can grant access to their home directory by giving other users any (or none) of the variety of permission levels that Windows offers. By default, your home directory can be accessed by you alone. Keep your directory structure in mind when assigning permissions. Other users must have at least Read permissions on all parent directories to the directory you are attempting to share. If you are using Windows, set permissions by a right-click on a file or folder, then choose Properties \| Security. If you are in DOS, help cacls gives information on modifying access control lists of files. This command works in a counter-intuitive manner and is not recommended. It is extremely easy to wipe out all permissions, so use it with caution. Warning: Never give Full Control to anyone for your home directory, or any subdirectory; grant access privileges only as needed. Full details on protecting your files and directories through permissions is here. Tokens and "Permission Denied" Messages: To write a file or to read a protected file in Windows, you must have an authentication "token" issued by Windows's security system. A token is normally issued to you when you log in. This token identifies you as the username you used to connect to the file server, and therefore you inherit all permissions of that username. Note: This is not a kerberos token. Backups: Every night all home directories are backed up. Scratch Disk Space: Each of Velocity's batch machines has at least 18 Gigabytes of scratch disk space mapped to its t:\ directory. For details see CTC Computing Resources. This space is locally attached to each machine for faster access. You may use this temporary scratch disk space as work space for your jobs while they are running. When your job ends, copy all the data that you wish to keep from t:\ to some other location, such as your home directory. Scratch space warnings: Files in scratch directories are not backed up. Currently files in t:\ on batch machines are not erased; you should explicitly delete sensitive files when your job ends. It is good housekeeping practice to delete all files from your t:\directory before releasing the machine(s) assigned to your job. The MCP Utility: The mcp utility in H:\CTC Tools is used to copy a file to local disk on every machine that your parallel job is using. For more information on the mcp command, see the mcp page. 3. Running Programs Paths: There are two paths in Windows: a system path and a user path. The system path is shared by everyone who logs into that computer and can only be modified by someone with administrator permissions. The user path is unique to each user account that logs into that computer, and can be modified by the account owner. The system path and the user path are concatenated to form the search path. The system path comes first, followed by the individual path and it is searched in that order. The search of the path stops as soon as the command is found. However you choose to set your path, remember that the names are case sensitive. In Windows, set the user path by left-clicking the system icon (in the control panel) and choosing the "Advanced" tab, then click on "Environment Variables" to display both the system variables and the user environment variables, including paths. You should be able to create a user path or select and edit an existing user path. Changes made to the path are in effect immediately, but currently running applications will not know about the changes. If you prefer to use command line prompts, type set path=%path%; (new paths to be appended). This will effect the current session only, which means that once this session is closed, the new path setting is gone. If there are two Command Prompts open, setting new path in one prompt will not effect the other. If you ssh to one of the login machines, and path settings are set in .bashrc files, the settings effect all the login ssh sessions under your account. Applications Software, Compilers and Programming Tools: A complete list of the software available on the login machines, various types of compute machines and the Collaboratory is updated as new packages are added. Programs can be compiled on your Windows machine, in the Collaboratory in Rhodes Hall, or on one of the login machines reached via a Remote Desktop Connection or ssh. For more information on writing programs, see Compiling and Porting Applications. Parallel Programming: There are a variety of ways to do parallel programming on Velocity: Native multithreaded applications - SMP OpenMP can use SMP nodes MPIPro is a way to use multiple nodes via message passing MPIPro and Open MP together can combine smp and message passing MPIPro can be used to run multiple processes on an SMP node and across multiple nodes. It automatically does message passing through shared memory on an SMP node and over the network between nodes. Parallel tools are available such as MPIPro from MPI Software Technologies and OpenMP from Intel. Running in Batch: Velocity Scheduler (vsched) is a job scheduler that distributes jobs to a cluster of workstations and/or to processors of a multi-processor machine. The scheduler decides when and how a batch job is run based on preferences set up by the user and system administrator(s). You can communicate with vsched using a few simple commands. Jobs are submitted to the batch system via an .xml file. When a job is submitted, it is held in the queue until the scheduler can allocate the required resources to run the job. If the job is a batch job, then the script is executed immediately after resources have been allocated. Once the job has completed, the user will (optionally) be notified. If the job is an interactive job, then the user must log in to the machine(s) when the job has started, and execute his programs. You cannot specify which machines to run on, but you may specify your requirements; vsched will then choose appropriate machines. Additional vsched documentation is available, including sample batch files and utilities. 4. Finding More Information WWW pages: CTC's collection of web pages contains extensive documentation for all aspects of the system. If the menus do not lead you to the topic you are seeking, try the Search button at the top of the page. Status: System status, including timely information such as system availability, schedule changes, and status of upgrades is available at http://www.tc.cornell.edu/Services/Support/general_topics.cluster_status.info.htm Help: DOS: For a list of DOS commands, type Help at the command line. For help on any one command, type help command_name. Windows: For help in Windows, choose Help from the Start menu. Menu bars: Most menu bars include a Help button. Cygwin: http://cygwin.com/faq/faq_toc.html Consulting: If you have questions or encounter difficulties while using any CTC system, CTC consultants may be able to help. Please send electronic mail to consult or call (607) 254-8686 and press 2 to speak to a consultant. Workshops offer intensive, hands-on instruction in the use of Theory Center resources. A diverse schedule of workshops offered in person at the Theory Center is supplemented by Virtual Workshops offered on the Internet.



© 2008 Cornell University \| Cornell University Center for Advanced Computing \| Copyright Statement