Responsibilities and Acceptable Use

The HBS research computing environment, composed of a number of servers and services, is a large, shared resource that is often at significant capacity and stores confidential and senstive research data. The compute jobs are isolated as much as possible by the LSF scheduler, OS sandboxing (CGROUPS), and identity management permissions. However, there are a number of things to keep in mind while using this shared resource so that the system can work as well as possible for everyone:

HBSGrid Compute Cluster

  • Ensure that you've familiarized yourself with our documentation about the cluster and have attended the required training(s). Note that if you did not take the required trainings in the required time frame, your access may be revoked. The time invested will more than make up for the time saved.

  • As they are the gateway to the cluster for everyone, don't run anything on the login nodes. Please use the drop-down menus, wrapper commands, or LSF job commands to have your work performed on the back-end compute nodes.

  • Be as accurate as possible when requesting memory or CPU cores for your jobs. Access to this shared resource depends on everyone honoring "Take what you need, and Need what you take." For more information, see https://grid.rcs.hbs.org/choosing-your-resources.

  • Ensure that you routinely monitor your interactive and batch jobs, so that you can terminate them if they stall or become idle. Remember that resources 'in use' are not available to anyone else.

  • Please do scaling tests if planning large numbers of jobs or parallel/multicore jobs to confirm that your code is correct and core usage is appropriate. Scheduling and then killing hundreds or thousands of jobs due to mistakes or problems is disruptive. See our documentation on parallelization at https://grid.rcs.hbs.org/parallel-processing.

  • Heavy I/O should be done on /export/scratch storage or storage local to the compute node. Heavy reading/writing large numbers of files while also running large numbers of (simultaneous) jobs can adversely affect others users' home and project folders. Please contact RCS if you have any questions.

  • Poorly behaved jobs or ones in the wrong queue may be terminated. Problem jobs or those not scheduled correctly can inhibit access and processing for other users, and may result in lost work. Thus, if a job is running improperly (e.g. excessive I/O; unreserved, excessive CPU usage) the job may be terminated. Repeated problems may result in disabling of a user's account.

  • Again, usage is intended only for legitimate purposes which benefit the research at Harvard University.

STORAGE

  • Do not store backups of your desktop or laptop data on the research storage, as this high-capacity, high-availability storage is expensive. Please use the HBS-supplied Code42 for backup purposes or a free cloud service (e.g., CrashPlan) if your data is not considered HRCI Level 3 or higher, or you might consider purchasing an external hard drive from your local electronics vendor.

  • Attend to any automated messages concerning "Out of Space" or "Over Quota" problems. These are not spam, and if ignored, may result in loss of data, for which neither RCS nor HBS IT is responsible.

  • Keep your project spaces clean and lean. See our document on archiving data within home folders or project spaces for more information (http://grid.rcs.hbs.org/archive-your-research-files).

  • Do not store data in directories not approved for the data classification level. For data with HRCI classifications higher than level 3, home directories and regular projects folders are not appropriate, and such data needs to be stored is a special location. Please contact RCS if you have any questions.

  • Heavy I/O should be done on /export/scratch storage or storage local to the compute node. Heavy reading/writing large numbers of files while also running large numbers of (simultaneous) jobs can adversely affect others users' home and project folders. Please contact RCS if you have any questions.

MARIADB

  • Store data with HRCI classifications in databases only if approved for that classification level. Please contact RCS if you have any questions or concerns.

  • For large or complex queries, please the EXPLAIN term or try to scale your work appropriately so that you do not adversely affect other MariaDB users.

  • Please see our documentation on MariaDB best practices (https://grid.rcs.hbs.org/db-best-practices) for information on how to make the best use of resources.

SOFTWARE

  • Software licensed for you via HBS or on the compute cluster is for your use only. Please do not share installers or license codes or in any way allow others to have access. And especially do not share your account credentials (username and/or password) to do so.

  • Don't install or run software that you do not trust or cannot verify the authenticity of the source. Since our work computers and research computnig environment houses confidential and sensitive research data, we must reduce the risk of exposure or loss. Know and trust where you obtained software and know thoroughly what the software does.

Updated 5/5/2020