Quiz

Use this self-check to test your understanding of data handling, paths, and storage quotas on the CoSTAR cluster.

These questions are browser-side self-checks. They are not formally graded and can be retaken as many times as needed.

--- primary_color: "var(--color-brand-primary)" secondary_color: "var(--color-background-secondary)" text_color: "var(--color-foreground-primary)" shuffle_questions: false shuffle_answers: true --- ### Which of the following statements about the CoSTAR storage systems are TRUE? Select all that apply: - [x] The personal home directory (``/users/<username>``) has a strict 30 GB storage quota. > Correct. Home directories are small and meant exclusively for code, configuration files, and basic environments. - [x] The WekaFS parallel storage workspace is built for ultra-fast performance but is not backed up. > Correct. High-performance filesystems bypass heavy backup overhead to maximize raw file I/O speed. - [x] There is no rigid, restrictive user-specific quota on the shared parallel scratch workspace, making it ideal for massive datasets. > Correct. The shared scratch space allows flexibility for large training sets, but it is a shared resource that must be managed responsibly. - [ ] The parallel scratch filesystem is fully backed up nightly by system administration. > Incorrect. Scratch storage is temporary and never backed up. You are responsible for preserving critical data. - [ ] Running heavy AI training loops directly out of your home directory is recommended for stability. > Incorrect. Home storage lacks the parallel throughput of WekaFS; doing this will bottle-neck your GPUs and quickly exhaust your 30 GB quota. ### What is the primary purpose of the high-performance parallel scratch workspace on CoSTAR? 1. [x] Running active compute jobs and handling large, high-throughput files during execution. > Correct. The parallel NVMe flash fabric is optimized to stream massive datasets directly to training models without I/O choke points. 1. [ ] Storing personal scripts, repositories, and local environment files permanently. > Incorrect. Long-term static files belong in your home space or project environments. 1. [ ] Long-term archival and backup of historical research data. > Incorrect. Scratch directories are unbacked and subject to periodic clean-up cycles. 1. [ ] Serving as a public file-sharing workspace for collaborators outside the university network. > Incorrect. Access is strictly confined within the cluster security perimeter. ### Which command-line tool is highly recommended for transferring large project directories or resuming interrupted file transfers to CoSTAR? 1. [x] ``rsync`` > Correct. ``rsync`` is highly efficient because it compares file deltas and can securely resume an interrupted transfer right where it left off. 1. [ ] ``scp`` > Incorrect. While useful for quick single-file drops, ``scp`` is inefficient for multi-gigabyte datasets and cannot cleanly resume broken transfers. 1. [ ] ``ftp`` > Incorrect. Plain FTP is an insecure, unencrypted legacy protocol and is completely blocked on the gateway. 1. [ ] ``cp`` > Incorrect. The ``cp`` command is exclusively for duplicating files locally within the cluster filesystem, not over the network. ### What is the storage quota for a user's personal home directory space on the CoSTAR login subsystem? 1. [x] 30 GB > Correct. Your personal home environment has a hard limit of 30 GB to preserve system stability on the head node. 1. [ ] 200 GB > Incorrect. This was a limit on legacy configurations; CoSTAR utilizes a streamlined 30 GB layout for configuration and code. 1. [ ] 750 TB > Incorrect. 750 TB represents the total pooled capacity of the flash-backed WekaFS array, not individual home allocations. 1. [ ] Unlimited > Incorrect. Cluster resources are finite and governed to ensure high availability for all users. ### Which of the following are considered best practices for managing storage on CoSTAR? Select all that apply: - [x] Regularly synchronizing important source files, models, and results to external platforms like GitLab or Hugging Face. > Correct. Because cluster volumes are not long-term archives, pushing critical assets to secure remotes prevents data loss. - [x] Cleaning up temporary files and clearing out old checkpoint runs from the parallel workspace as soon as a job series finishes. > Correct. Keeping the parallel volume clear helps maintain top speed and ensures ample room for other active research teams. - [ ] Running all computational tasks out of your home directory to ensure they remain safe from cleanup loops. > Incorrect. Tasks should always be executed within the parallel workspace to leverage proper storage bandwidth. - [ ] Storing all model training data inside scratch directories indefinitely so you never have to re-download them. > Incorrect. Treat scratch purely as an active runtime workspace. Data left idle can be flagged or automatically purged during maintenance windows.