From Documentation
Revision as of 17:12, 21 January 2013 by Bge (Talk | contribs) (Handling of allocations at SHARCNET)

Jump to: navigation, search

Different problems require differing degrees of computational resources to solve. In order to effectively meet the needs of demanding computational problems SHARCNET participates in a national resource allocation competition on behalf of Compute Canada, as well as operating a continuous competition for short-duration resource allocations throughout the year.

These competitions allow researchers (and their groups) to obtain high-priority job queues on SHARCNET systems so that their jobs start more quickly than regular jobs. It also provides access to storage resources beyond the default limitations (both duration and size).

Compute Canada NRAC (National Resource Allocation Competition)

The Compute Canada NRAC allocation process accepts applications for substantial dedicated resource allocations every September, as well as other specialized allocation rounds throughout the year.

If you have a Compute Canada account please log in to the Compute Canada website and visit your resource applications page for further details.

SHARCNET Continuous Small Dedicated Resources

The SHARCNET Continuous Small Dedicated Resources process accepts applications for substantial resource allocations throughout the year. The process is initiated by contacting a SHARCNET staff member to discuss the computational problem at hand and the suitability of it for the program.

For more information, please refer to the dedicated resources application guidelines for the current (2012) round.

Handling of Allocations at SHARCNET

Project Membership

NRAC and CSDR allocations, also conventionally termed NRAC projects (NRAPs) or SDR projects at SHARCNET, are handled on a per-group basis. Unlike the UNIX group of users sponsored by the PI, this is an ad hoc group formed only for the dedicated resource allocations. This group includes all users including those collaborators who are sponsored by other PI(s) as well. This extended group defines a new UNIX group for each project, with the name, e.g. nrapnnn or sdrnnn, pertinent to the dedicated resource allocations NRAC nnn or SDR nnn, for exclusive access to the dedicated queue and storage.

By default, users sponsored by the same PI shall have the group membership for this dedicated resource allocation project group. The PI may add or remove the membership of a user through the Compute Canada web portal or simply ask SHARCNET to to do so.

Dedicated Queues

In addition to this new unix group, special queues corresponding to the project are set up on the designated systems which grant members the ability to submit high-priority jobs will start before regular jobs. The queue name will be of the following form

NRAP_nnn for NRAC allocation nnn,

SDR_nnn or DR_nnn for small DR or DR allocations.

The size of the queue (both number of jobs and number of concurrent cpu cores) is specified at the outset of the project based on it's needs.

Dedicated Storage

If a group applied for a storage allocation they will typically be given a new directory for the project on the required filesystem, with any storage quotas and expiry timescales set at the outset of the project based on it's needs.

Each group will have access to

/work/group and

/scratch/group

on global work and system scratch respectively. The size of the storage of these directories is set to the dedicated storage awarded or the default.

Running Jobs

The following are a number of highlighted points that will help you to get started with using the dedicated resources:

1. USING DEDICATED QUEUE To submit a job, use the following command

   sqsub -q dedicated_queue -r runtime -o output ./yourprog [ yourprog_arg_list ]

where runtime is estimated runtime in the format of [m|h|d], e.g. 50m for 50 minutes, 3.5h for 3.5 hours and 4d for 4 days, etc. Your job will be terminated if the run exceeds the run time estimate. The maximum runtime for each job is seven days. The option -o output specifies the name of the file output that you would like to have all output from standard out to be redirected to. For the details of sqsub options, use command

   sqsub -h

2. RUNNING PARALLEL JOBS If you are running parallel jobs, i.e. MPI or threaded jobs, use the following command

   sqsub -q dedicated_queue -f mpi|threaded -n num_cores -r runtime -o output ./yourprog [ yourprog_arg_list ]

Note the option -f mpi for MPI jobs or -f threaded jobs MUST be used in conjunction with the dedicated queue.

3. CHECKPOINT To prevent unexpected system outage, you should enable _check-point_ in your applications, so your computational time won't be wasted in the event of a system outage.

4. QUEUE PRIORITY The dedicated queue has a higher priority than regular queues on the designated system, therefore, your jobs submitted to the dedicated queue have a better chance to start faster. It should not be expected, however, that dedicated jobs will be started immediately. It is common that dedicated jobs also have to wait for some time before they get started.

5. ACCESS PRIVILEGE Access to the dedicate queue and storage is restricted to the users listed in this project only. It shall not be confused by the access privilege of the user group sponsored by the PI. Access to the dedicated queue and storage is by project, not the user group. For example, a user sponsored by PI but not listed as a project member will NOT have the access to queue and the storage.

6. FAIRSHARE Your usage of dedicated resources will not affect your group fair share for using regular queues. It is important to remember always using the dedicated queue to run your dedicated resource allocation jobs. The progress, usage, and job submission history (in graphs) of your dedicated resource project is monitored and updated hourly.

Project Log

Groups can see the projects associated with their NRAC or CSDR allocations in the web portal here. All communication with SHARCNET concerning your allocation should be conducted through this interface when possible. It also includes information on how well one is progressing towards the completion of their allocation and other supporting information.