This guide is a work in progress and is not complete yet. We are actively working on adding more content and information.
Welcome to Sherlock!#
Sherlock is a High-Performance Computing (HPC) cluster, operated by the Stanford Research Computing Center to provide computing resources to the Stanford community at large. You'll find all the documentation, tips, FAQs and information about Sherlock among these pages.
Why use Sherlock?#
Using Sherlock for your work provides many advantages over individual solutions: hosted in an on-premises, state-of-the-art datacenter, the Sherlock cluster is powered and cooled by installations that are optimized for scientific computing.
On Sherlock, simulations and workloads benefit from performance levels that only large scale HPC systems can offer: high-performance I/O infrastructure, petabytes of storage, large variety of hardware configurations, GPU accelerators, centralized system administration and management provided by the Stanford Research Computing Center (SRCC).
Such features are not easily accessible at the departmental level, and often require both significant initial investments and recurring costs. Joining Sherlock allows researchers and faculty members to avoid those costs and benefit from economies of scale, as well as to access larger, professionally managed computing resources that what would not be available on an individual or even departmental basis.
How much does it cost?#
Sherlock is free to use for anyone doing sponsored research at Stanford. Any faculty member can request access for research purposes, and get an account with a base storage allocation and unlimited compute time on the global, shared pool of resources.
No CPU.hour charge
Unlike all Cloud Service Providers and many HPC systems, there is no usage charge on Sherlock.
When you submit your work on Sherlock, you don't need to keep an eye on the clock and worry about how much that run will cost you. There is no limit on the total amount of computing you can run on the cluster, as long as resources are available, and there's no charge to use them, no matter how large or small your computations are.
In case those free resources are not sufficient, Stanford Research Computing offers Faculty members the opportunity to invest into the cluster, and get access to additional computing resources for their research teams. Using a traditional compute cluster condominium model, participating faculty and their teams get priority access to the resources they purchase. When they're idle, those resources are available to use by other owners on the cluster, giving them access to virtually unlimited resources.
Searching the docs
If you're looking for information on a specific topic, the Search feature of this site will allow you to quickly find the page you're looking for. Just press S or F to open the Search bar and start typing.
To help users take their first steps on Sherlock, we provide documentation and information through various channels:
You are here
|www.sherlock.stanford.edu/docs||information to help new users start on Sherlock, and more in-depth documentation for users already familiar with the environment.|
|Changelog||news.sherlock.stanford.edu||announces, news and updates about Sherlock.|
|Dashboard||status.sherlock.stanford.edu||status of Sherlock's main components and services, outages, maintenances.|
To get started, you can take a look at the concepts and glossary pages to get familiar with the terminology used throughout the documentation pages. Then, we recommend going through the following sections:
Acknowledgment / citation#
It is important and expected that publications resulting from computations performed on Sherlock acknowledge this. The following wording is suggested:
Some of the computing for this project was performed on the Sherlock cluster. We would like to thank Stanford University and the Stanford Research Computing Center for providing computational resources and support that contributed to these research results.
Research Computing support can be reached by sending an email to email@example.com and mentioning Sherlock.
How to submit effective support requests
To ensure a timely and relevant response, please make sure to include some additional details, such as job ids, commands executed and error messages received, so we can help you better. For more details, see the Troubleshooting page.
As a member of the Sherlock community, you're also automatically subscribed to the sherlock-announce mailing-list, which is only used by the SRCC team to send important announcements about Sherlock,
We offer Sherlock onboarding sessions for new uers every first Wednesday of the month 1-2pm PST.
This one hour session will be a brief introduction to Sherlock's layout, scheduler limits, file systems and some job submission/software installation best practices for new users. So, come by if you are new to Sherlock and/or HPC in general.
Onboarding session Zoom link
Sending a question to firstname.lastname@example.org is always the best first option for questions. That way you can include detailed descriptions of the problem or question, valuable output and error messages and any steps you took when you encountered your error. Also, everyone on our team will see your ticket, enabling the most appropriate group member to respond.
Office hours are a good place for more generalized questions about Sherlock, Slurm, Linux usage, data storage, queue structures/scheduling, job optimization and general capabilities of Sherlock. It's also useful for more technically nuanced questions that may not be easily answered with our ticketing system. In office hours some problems can indeed be solved quickly or progress can be made so that you can then work self-sufficiently towards a solution on your own.
We'll be holding remote office hours via Zoom, for the time being.
Office hours times
Click here to join the Sherlock Office Hours Zoom
- Tuesday 10-11am
- Thursday 3-4pm
You'll need a full-service SUNet ID (basically, a @stanford.edu email address) in order to join Office Hours via Zoom. Make sure to authenticate at (url_zoom) If you do not have a full service account, please contact us at email@example.com.
If you can't make any of the Office Hours sessions, you can also make an appointment with Sherlock's support team.
What to expect#
We cannot accomodate walk-ins: we're unfortunately not staffed to accommodate walk-ins, so please make sure that you're planning to stop by during office hours. We will not be able to help you otherwise.
We can rarely help with application-specific or algorithm problems.
You should plan your projects sufficiently in advance and not come to office hours at the last minute before a deadline. Sherlock is a busy resource with several thousand users and you should not expect your jobs to complete before a given date.
Not all questions and problems can be answered or solved during office hours, especially ones involving hardware, filesystem or network issues. Sherlock features several thousand computing, networking and storage components, that are constantly being monitored by our team. You can be sure that when Sherlock has an issue, we are aware of it and working on it.
Sherlock is present on the Stanford Slack Grid, and you're more than welcome to join the following channels:
#sherlock-announce, for announcements related to Sherlock and its surrounding services,
#sherlock-users, as a place for Sherlock users to connect directly with each other. If you have a general question about software used on Sherlock, want to reach out to other Sherlock users to share tips, good practices, tutorials or other info, please feel free to do so there.
For more details about the SRCC Slack Workspace, and instructions on how to join this workspace and its channels, please see https://srcc.stanford.edu/support.
Slack is not an official support channel
Please note that while SRCC staff will monitor these channels, the official way to get support is still to email us at firstname.lastname@example.org.
If you're in a rush1, here's a 3-step ultra-quick start:
- connect to Sherlock
$ ssh login.sherlock.stanford.edu
- get an interactive session on a compute node
[kilian@sh-ln01 login! ~]$ sdev
- run a command
[kilian@sh02-01n58 ~]$ module load python [kilian@sh02-01n58 ~]$ python -c "print('Hello Sherlock')" Hello Sherlock
Congrats! You ran your first job on Sherlock!
Here's what it looks like in motion: