UBC ARC Sockeye FAQs

HPC, or High Performance Computing, is the ability to process and analyze complex data at very high speeds. These computations are performed on HPC clusters, which are made up of many networked compute nodes, each containing CPU and/or GPU cores. HPC clusters, including UBC ARC Sockeye (or simply “Sockeye”), offer several thousand times the computing power of standard desktop computers and are useful for research that involves large datasets or complex simulations that would otherwise take unreasonably long on personal machines. 

Alliance and Sockeye 

UBC ARC operates at the institutional layer of Canada’s digital research infrastructure (DRI) ecosystem. We provide local HPC resources like Sockeye, along with support and training tailored to UBC researchers. We complement the national layer by acting as a bridge to Alliance systems, helping researchers to apply for Alliance resources and troubleshoot issues and scale their work beyond UBC. 

The Digital Research Alliance of Canada (or simply “Alliance”) operates multiple HPC clusters across Canada (e.g., Fir, Rorqual, Nibi). Like Sockeye, these systems run on Linux and use Slurm scheduler to manage compute workloads. 

If you have questions about which system will better support your research needs or how to use both effectively, please reach out to us at arc.support@ubc.ca so we can set up a consultation.  

Access to Sockeye requires an allocation. Allocations are awarded to one Allocation Owner – typically a UBC faculty member or researcher with Research Spending Authority. The Allocation Owner or Designated Contact may then add users to their allocation, such as students, postdoctoral fellows, or research staff, provided they have a valid UBC CWL and an institutional email address. External collaborators can be added if the Allocation Owner sponsors them for a Guest CWL. 

To apply, complete the Sockeye Application Form. If you have questions about eligibility or the application process, contact arc.support@ubc.ca. 

Researchers with large‑scale needs may also apply for national resources through the Digital Research Alliance of Canada. ARC staff can advise on how to use both systems effectively. 

Every Sockeye allocation has three types of participants: 

Allocation Owner 
The faculty member who applies for the allocation. The Allocation Owner is responsible for the allocation, its renewal, and compliance with ARC policies. 

Designated Contact 
An optional additional faculty or staff member identified by the Allocation Owner in the application form. The Designated Contact may be contacted by ARC regarding the allocation, especially for administrative matters, renewals, or user management. The Allocation Owner remains accountable for the allocation, but the Designated Contact helps share responsibility for communication and coordination. 

Allocation Users 
Students, postdoctoral fellows, research staff, or collaborators added by the Allocation Owner. Allocation Users must have a UBC CWL and an institutional email. They can run jobs and access the allocation’s compute and storage resources but cannot manage or renew the allocation themselves. 

Allocations expire on June 30 each year. Renewal reminders are sent before this date. 

If renewed, access continues seamlessly. If not renewed, compute access ends immediately and all users in the allocation lose access to the scheduler. A 14‑day grace period is provided to copy or move data, but no new jobs can be run. Allocations may be reactivated by submitting a renewal form. In some cases, extensions to retrieve data can be arranged by contacting ARC support. 

Sockeye consists of over four hundred CPU compute nodes and fifty GPU nodes, providing 15,872 CPU cores and 200 NVIDIA Tesla V100 GPUs. Compute nodes are available in several configurations, ranging from base nodes with 192 GB RAM to large nodes with up to 768 GB RAM. GPU nodes are equipped with either 16 GB or 32 GB V100s. 

All nodes are connected by a Mellanox InfiniBand EDR interconnect with 100 Gbps bandwidth. Sockeye’s storage system includes a 1.5 PB Lustre parallel filesystem for high‑performance workloads and 1.2 PB of Isilon storage across two performance tiers. 

Full details are available on the Sockeye Technical Specifications page. 

Each Sockeye allocation comes with a set of default compute and storage resources. The table below summarizes the standard quotas and policies associated with each type of resource available. These defaults apply to most allocations, though adjustment can sometimes be made based on specific research needs. 

If additional storage or compute capacity is needed for a short period, researchers may contact ARC with details of their request, including what is needed, why it is required, and the deadline they are working toward. These requests are granted only on a time‑limited basis to address urgent research needs such as conference submissions, paper deadlines, or thesis defenses. 

Researchers may also occasionally ask for a priority bump, allowing a specific job to move up in the queue for time‑sensitive needs. Priority adjustments apply only to individual jobs, do not affect future scheduling, and do not change Fairshare. They should not be used routinely or to bypass the standard scheduling policy. 

ARC OnDemand is a browser‑based portal that allows researchers to use Sockeye without relying on the command line. Through OnDemand, users can log in with their CWL and access the cluster interactively. The service provides a simple way to run jobs, transfer files, and launch applications such as Jupyter notebooks or RStudio directly from a web browser. 

OnDemand is especially useful for new users, and researchers who prefer a graphical interface. It connects directly to the same resources available through the command line, including GPU nodes, but is limited to specific partitions configured for interactive use. 

For details and login instructions, see the ARC OnDemand page. 

Sockeye storage is designed to support active research workflows through the home, project, and scratch spaces provided in each allocation. Details of the default quotas are listed in the 'How are the resources allocated?' section above. For detailed list of Sockeye’s storage volume, refer to Storage Volume. 

Project space should be used for persistent research data that needs to be shared within the allocation, while scratch space (also shared between all members of an allocation) is intended for temporary working files and may be purged as needed to maintain availability. Home directories are private to each user and best suited for scripts and configuration files. 

For long‑term retention or larger unstructured datasets, researchers should use UBC ARC Chinook, UBC ARC’s dedicated object storage platform, which integrates with Sockeye through Globus for high‑speed transfers. 

Common research software is available through modules. The current catalogue is maintained on the ARC software page. Researchers may also install software independently within their allocation. 

If additional software is needed, Allocation Owners may request installation through the ARC software request form. ARC staff review requests for technical compatibility and licensing requirements. 

Interactive tools such as Jupyter notebooks and RStudio are supported through ARC OnDemand. 

Sockeye supports high‑speed data transfers through Globus. Globus allows researchers to move large datasets reliably between Sockeye and other endpoints, including local machines and UBC ARC Chinook. 

Researchers can install Globus Connect Personal on their desktop or laptop to create a personal endpoint, enabling direct transfers between Sockeye and their local system. 

Maintenance Windows 
Sockeye is subject to a regularly scheduled maintenance window on the third Tuesday of every quarter. Maintenance usually runs from 09:00 to 18:00 PT but may be shorter or longer depending on the need. Regular maintenance is required to maintain service reliability and security and introduce new features and capabilities. You can check scheduled maintenance windows, real‑time outage reports, and notes on performance issues on the UBC IT Status page under Advanced Research Computing. 

User Impact 
During a maintenance window, Sockeye users cannot log in or run jobs. The scheduler will not start any job with a walltime that overlaps with the maintenance period; shorter jobs scheduled to finish before maintenance begins will run normally. All queued jobs will be held until maintenance ends. In rare cases, jobs running before the window may need to be resubmitted. Users are encouraged to checkpoint or complete their work ahead of time and to review their job queue once maintenance is complete. 

System Status 
Up‑to‑date information about Sockeye and Chinook is posted on the UBC IT Status page under Advanced Research Computing. This includes scheduled maintenance windows, real‑time outage reports, and notes on performance issues. Users are encouraged to check the status page before reporting an issue, especially during maintenance windows, as jobs and data transfers may be paused or delayed. ARC also posts maintenance reminders through the Sockeye user mailing list, the Message of the Day (MOTD) displayed at login, and the Technical User Documentation site. 

User Notices 
ARC communicates with Sockeye users about system status and changes through multiple channels: the ARC‑SOCKEYE‑USERS mailing list, the Sockeye Message of the Day (MOTD), and notices posted on the UBC ARC Technical User Documentation. 

Mailing List 
All Sockeye users are automatically subscribed to the ARC‑SOCKEYE‑USERS mailing list. This list is the primary way to stay up to date on Sockeye, including announcements about maintenance, downtime, policy changes, new services, and training opportunities. Users may freely subscribe or unsubscribe through the UBC Mailing List platform. 

The mailing list archive is also publicly viewable: 

• Sockeye: arc‑sockeye‑users@lists.ubc.ca 

UBC ARC-led training 

The Training and Resources page on the UBC ARC website maintain a list of historical and upcoming research computing training opportunities. 

The Alliance 

The Alliance offers training via regional partners including SHARCNET, SciNet, Calcul Quebec, among others. A curated list of training resources is available here: Alliance Training Resources 

Acknowledging UBC Advanced Research Computing (ARC) is important for demonstrating the impact of Sockeye and related services. These acknowledgments help ARC secure funding, maintain infrastructure, and continue supporting the UBC research community. 

Researchers should acknowledge ARC in any publication, presentation, report, or proposal where ARC systems (such as Sockeye, or Chinook) or ARC staff expertise have contributed. 

General Acknowledgment 

This research was supported in part through computational resources and services provided by Advanced Research Computing at the University of British Columbia. 

Sockeye Citation 
If you are citing Sockeye directly, please use the official DOI reference in your preferred citation style: 

UBC Advanced Research Computing, UBC ARC Sockeye. University of British Columbia, [year]. doi:10.14288/SOCKEYE 

Acknowledgment of ARC Staff 
If ARC staff provided significant support or guidance, you may also acknowledge them by name: 

The authors thank [insert name(s)] from Advanced Research Computing at the University of British Columbia for their support in this work. 

Publication Showcase 
ARC maintains a collection of publications from UBC researchers who have acknowledged our services. If you would like your work featured, please submit it through the ARC Publications page.