Use cases for Singularity...

[Gregory M. Kurtzer]

Hi everyone,

I need your help to put together a list of use-cases of how Singularity is being used, both for science and otherwise (e.g. Dolmades). The more info I can obtain on this, the better. Please feel free to respond either publicly or private to this email/thread (to help me keep track of responses) and let me know how you are using Singularity!

Thank you!

--

Gregory M. Kurtzer

CEO, SingularityWare, LLC.

Senior Architect, RStor

Computational Science Advisor, Lawrence Berkeley National Laboratory

[Paolo Di Tommaso]

Hi Greg, 

Nextflow uses Singularity to deploy large-scale distributed scientific workflow, mostly genomic pipelines, across cluster and the cloud in portable manner. It's used regularly at the Center for Genomic Regulation (CRG) and many other research centers and universities such as Pasteur Institute (France), SciLifeLab (Sweden), Sanger Institute (UK) between the other. 

Well, you should know this blog post.   

Best,
Paolo

--
You received this message because you are subscribed to the Google Groups "singularity" group.
To unsubscribe from this group and stop receiving emails from it, send an email to singularity+unsubscribe@lbl.gov.

[Shashi Ranjan]

Can any body explain me difference between singularity and docker in detail?

With Regards,

Shashi Ranjan,
Mobile # 9008189000
in.linkedin.com/in/shashiranjanIndia

To unsubscribe from this group and stop receiving emails from it, send an email to singularity...@lbl.gov.

[Paolo Di Tommaso]

There's a full scientific paper for that 

http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0177459

Enjoy,

Paolo

[Mats Rynge]

> I need your help to put together a list of use-cases of how Singularity is being
> used

Greg,

Open Science Grid [1] is using Singularity to provide a consistent environment, and to allow for our users to bring in custom images, for high throughput jobs across contributing compute sites. OSG VO [2] (a subset of Open Science Grid) is consistently running ~300k Singularity instances per day. See attached graph. Other projects on OSG are using Singularity as well.

OSG VO users can put requirements on their jobs to make them go to sites which have Singularity installed, and specify in the job definition what image to wrap the job with [3]. However, most of the instances are users running regular jobs under a default image. Many of these users do not even know they are running inside Singularity - all they see is a consistent environment!

Also note how images are distributed in an expanded file system form over CVMFS [4], which allows us to efficiently access the images from anywhere across our distributed system. Only the accessed parts of the images has be transferred, and with the aggressive caching of CVMFS, even running the amount of instances we do have no measurable impact on the network. Currently images are synced from Docker Hub, but we will be exploring direct Singularity imports with 2.3 if we can get it to work as regular users (not tested yet).

/Mats

[1] https://www.opensciencegrid.org/
[2] https://osgconnect.net/
[3] https://support.opensciencegrid.org/support/solutions/articles/12000024676-singularity-containers
[4] https://cernvm.cern.ch/portal/filesystem

[vanessa s]

I just have to comment... that is so awesome :)

Best,

Vanessa

> --
> You received this message because you are subscribed to the Google Groups "singularity" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to singularity...@lbl.gov.

> <osgvo-singularity-instances.png>

[Brian Bockelman]

Hi Greg,

The CMS (https://cms.cern/) experiment additionally uses Singularity to help provide jobs with a consistent runtime environment across a variety of sites worldwide.  Currently, about 15 have enabled Singularity support: the overall number is on the increase.

CMS and OSG developed the use case jointly, so I don't have much to add on top of Mats' very good description.

The corresponding "job count" graph is below.  So, adding together CMS and OSG, there's probably about 0.5M instances launched a day.

HTH,

Brian

[Rick Wagner]

Hey Greg,

I just presented at the Lustre User Group meeting about an automation platform driven by file system events. The use case we developed was user-defined metadata extraction using Singularity under the hood. Singularity worked well for this because the processes doing the extraction can all be done in user space, and users can manage a suite of portable images so the same extraction can happen on different systems.

https://lug17.sched.com/event/AMQD/responsive-storage

http://schd.ws/hosted_files/lug17/2f/Fri05-WagnerRick-LUG2017-Wagner-Chard-v2.pdf

—Rick

On Jun 9, 2017, at 6:08 AM, Brian Bockelman <bock...@gmail.com> wrote:

Hi Greg,

The CMS (https://cms.cern/) experiment additionally uses Singularity to help provide jobs with a consistent runtime environment across a variety of sites worldwide.  Currently, about 15 have enabled Singularity support: the overall number is on the increase.

CMS and OSG developed the use case jointly, so I don't have much to add on top of Mats' very good description.

The corresponding "job count" graph is below.  So, adding together CMS and OSG, there's probably about 0.5M instances launched a day.

HTH,

Brian

<CMS-singularity-count.png>

[Guillaume DOLLÉ]

Hi Greg,

We use singularity in the context of a science H2020 european project MSO4SC.
One of the goal is to deploy cloud based services for several mathematical frameworks (fenics, feel++, opm)
using container technologies (singularity, docker, ...) to provide a convenient way for fast prototyping, modeling and run large-scale simulations for (non) expert users.
Developers typically provides containers for their mathematical framework/applications and the final user
configure/run simulations via a web interface (singularity is used seemlessly, it's a w.i.p).

see http://mso4sc.elmerex.hu/ for more info.

Best,

Guillaume

[David Trudgian]

Hi,

The UT Southwestern BioHPC group (https://portal.biohpc.swmed.edu) is using Singularity to ease installation of applications on our HPC cluster. We are currently running RHEL 6.x (upgrading to 7.x later this year), and continue to have issues with various applications that expect newer distros / libs. Our user base of biomedical researchers are often not familiar with software installation, so we need to maintain a very large collection of software modules relative to our size and number of team members.

Singularity offers huge promise as a way to allow us to quickly deploy software that e.g. expects recent Ubuntu, without days of effort rebuilding from source, tracking down deps etc. Deep learning and visuzalization packages are particularly problematic for us. The effort required to install frequent updates, building from source on an older OS, is extremely high relative to the usage of the packages. However, the packages are vital to some researchers so we must make them available.

Example - we had a request from 3 users for the latest version of Blender (www.blender.org). Binaries require GLIBC 2.19 which is newer than our cluster has. We use a Windows VM approach to support some other software, with 3D rendering on the host GPU using the NICE DCV product. This didn't work for blender. A singularity image of the latest blender on Ubuntu took approx. 1 hr to produce. GPU support worked, as did using VirtualGL inside the container to render on the GPU, forwarding graphics into a TurboVNC server on the host for display on the user desktop. In this case Singularity saved multiple days of effort and frustration building the deep dependency tree of blender.

We also plan to leverage https://biocontainers.pro and singularity to provide trivial access for user to a very large range of versioned bioinformatics software, automatically making new packages/versions available. We have an in-house web-based workflow platform using Nextflow, and wish to encourage use of containerized apps here as a way of improving long-term reproducibility and insulating bioinformaticians from the impacts of cluster updates etc, which can impact a traditional module stack of software.

Finally, it's possible that Singularity will give users enough freedom that we can avoid having to implement complex multi-image cluster or openstack VM environment which has been looking increasingly likely.

Many thanks for all your work!

--

Dave Trudgian

Computational Scientist

UT Southwestern BioHPC