University of California

Getting Started Guide

The outline below describes, at a very high level, everything that needs to take place for a new user to begin running jobs in the ShaRCS environment. Skip over any steps that do not apply to you or which you have already completed, and click on the links of those steps for which you want more details. If you need assistance, please contact ShaRCS Help. Someone will reply as quickly as possible to address your questions and concerns.

How to get started using ShaRCS

  1. Get an allocation
  2. Access the system
    1. logging in
    2. one-time passwords
    3. login nodes
    4. how to connect
  3. Prepare your job
    1. compiling a new code
    2. porting an existing code
  4. Run your job
    1. job queue overview
    2. job submission options
  5. Data storage and job output
  6. Troubleshooting

If you have questions about any of the topics covered below or the associated detail pages, please contact Email Support. Be specific in your description of the question or problem, and provide error messages, log file output, unexpected behavior, job submission details, job ID numbers, and other relevant information so the support staff can focus on a rapid solution.

Allocations on ShaRCS

To use the ShaRCS resources, you must work for a project that has been given an allocation. During the initial pilot phase, all projects come from a predetermined list, and only those projects have access to the resources. Once the pilot phase has completed (projected to be early 2011), UCOP plans to make the ShaRCS resources available to a broader UC community. See the About ShaRCS section for more details.

If you work on a pilot project supported by ShaRCS, you will be able to run jobs using a pilot project account. User logins will be created for each authorized user on the project. These accounts will be managed by the Principal Investigator of the project in cooperation with the ShaRCS system administrators.

Access the System

All requests for user accounts on ShaRCS should go through the project PI (or a designated representative). This person will be the primary contact with ShaRCS administrators who create and manage user accounts. Users who are authorized to login and utilize the resources will receive a One-Time Password (OTP) device (or software tool, pending future availability) to generate passwords for use on the login nodes. OTP will be the only method for users to log in to ShaRCS. See the One-Time Password page for complete details.

Each project will have a designated login address representing one or several nodes in a round-robin DNS configuration. Each project will be assigned to log in through one address, and use either the North Cluster (Mako) or the South Cluster (Thresher) for all its ShaRCS computational services. In the pilot phase, projects will not have access to both clusters. See the Login page for the specifics of your project.

Prepare Your Job for Execution

You will need to compile codes specifically for use on ShaRCS. You can write new codes directly on ShaRCS or port existing codes from other HPC resources and recompile them to run on ShaRCS. See the Hardware Descriptions page to learn about the ShaRCS execution environment. See the Compiling page for help with compiling codes for ShaRCS.

Run Your Job

To run a job on ShaRCS, you can use interactive mode or batch mode. In either case, submit the job to one of the job queues and wait for the scheduler to allocate compute nodes based on your request. Batch jobs are written in the PBS script language, while interactive jobs use the msub -I command run from a login node.

Queues are specified by name in the job submission command or script. Each queue has different characteristics to support particular types of jobs. The currently available queues are named express, normal, short, and long. Other queues may be defined as usage dictates. You can run the qstat -q command to display the names of all available job queues.

The login nodes are to be used only to launch interactive jobs, and to prepare and briefly test batch jobs. Since all users share access to login nodes, production runs are not to be submitted for execution on them. Full runs on these nodes may fail to meet performance expectations, and may cause denial of service to other users. If you need assistance launching jobs to compute nodes, please request help from .

See the Running Jobs page for more information.

Data Storage and Job Output

Several locations are available for temporary storage of data associated with ShaRCS jobs. Long-term, permanent storage is not provided as a formal component of the resource. Your needs in this area should be directed to high-capacity storage facilities outside the ShaRCS cluster and its associated filesystems. All local storage on ShaRCS is subject to removal without warning after a reasonable time, is not backed up, and may be accidentally lost or removed without warning with no capability for recovery. See the File Storage page and Data Transfer page for more information.

Troubleshooting and Debugging

Debugging tools and other troubleshooting mechanisms are available on ShaRCS, and you may request help from the support staff to resolve problems associated with running your jobs and interpreting unexpected or unexplained behavior. Contact for more information.