Computing resources

Contents

Computing resources

SBND has a number of computing resources available, many of which are supplied by Fermilab Computing Sector (SCD when we talk about the Scientific Computing Division). Here we list some of them.

Fermilab Computing Division has produced a very short flyer explaining in a nutshell the different credentials used in Fermilab. It can be found as document 5892 in the SCD DocDB. It is a strongly recommended reading, to everybody!

SCD liaison

The Scientific Computing Division Liaison is a person from Fermilab SCD who acts as a bridge between SCD and an experiment. They is typically fairly informed of what is going on, computation-wise, in the experiment, and in the computing division [1]. They is also able to route any request to the proper people, but they is not necessarily required to do actual work on solving the merit of the issues. Also, they is not required to be a member of the experiment.

The current liaison for SBND is Vito Di Benedetto (SCD).

[1] This is defined as knowing what goes on in the computing division more than most experiment members, and knowing what is going on in the experiment more than most of the computing division people.

Computing Access

To get a Fermilab computing account follow the instructions at this site: https://get-connected.fnal.gov/accessandbadging/access/. Our experiment number is T-1053 (SBND).

Write down the initial Kerberos and Services Account passwords before submitting. They will be necessary once your accounts have been created.

You will receive an email once this is complete which gives you your Fermilab email account and username (you will need this for service now).

The next step is to “kerberize” your machine. You can follow instructions here:

Once your computer is SSH-ready and kerberized, you should be able to log on by typing:

kinit ${USER}@FNAL.GOV
# insert your kerberos password
ssh -Y ${USER}@sbndgpvm01.fnal.gov

(where $USER is your Fermilab user name).

More details can be found ta the Newbie Material wiki page

If this does not work, please submit a to request a SBND project account:

  1. ServiceNow -> Request something; search for something like: Experiment Computing Account and you’ll be offered a “Experiment/Project/Collaboration Computing Account” match, that you can pick;
  2. Enter your name, then select T-1053 (SBND)
  3. Verify that the correct Fermilab principal is displayed when you enter your name
  4. Enter your home institution, then click .

You can reset your Fermilab passwords for service-now here: https://password-reset.fnal.gov/showLogin.cc If you can’t remember your password because you didn’t write it down then you will have contact the service desk (+1 (630) 840-2345).

Where to work: interactive nodes (“GPVM”)

We have some dedicated “nodes” where we can log in and work interactively. These are General Purpose Virtual Machines (GPVM); they are shared by all SBND collaborators, and they can not be the fastest way to develop and run analyses. But they see all the resources that we have available: you can read data from SBND disks and tapes, and you can submit SBND jobs to the grid. The access is via SSH with FNAL.GOV Kerberos credentials.

The nodes and their OS version are:

Node OS
sbndgpvm01.fnal.gov SLF 7
sbndgpvm02.fnal.gov SLF 7
sbndgpvm03.fnal.gov SLF 7
sbndgpvm04.fnal.gov SLF 7
sbndgpvm-test-al9.fnal.gov Alma 9
sbndbuild01.fnal.gov SLF 7
sbndbuild02.fnal.gov SLF 7

The GPVM interactive machines have no relevant local storage, and areas in /exp/sbnd and /pnfs should be used (the home directory works too, but it has small allowance).

The interactive “build nodes” have roughly the same environment as the other GPVM’s. The policy is to use these machines only to build code and running tests; this excludes job submission and local running of job campaigns and of single jobs beyond the size of a test job (say, 100 events).

Information about all SBND GPVM’s is on a separate page.

Opening a ticket in Fermilab Service Desk

This is the favourite sport of ours: when something goes wrong, open a ticket. You want to open a Service Desk ticket if your job submission that used to work yesterday fails today, if the GPVM are very slow (if they are just slow, it’s normal), if you are denied access to some resource you expect to deserve. You don’t open a service desk ticket if your jobs crash (most of the time, you just have to fix them), if there is a bug in SBND code (e-mail to SBND mailing list, write about it on SBN Slack channel (#sbnd_general) and/or open a GitHub issue) or if you need help with using LArSoft (LArSoft wiki, LArSoft mailing list, and SBND mailing list).

You need to log in via Fermilab “Services” password (the same you use to access Fermilab e-mail and Redmine), and then you go at:

https://fermi.servicenowservices.com/

That will bring you to the “web portal” interface where you can generically “request something” and a human being will sort it out in due time. Another way is to access the older service catalog. Then, on the left, you’ll have a list of “Self service” items: selecting “Scientific Computing Services” or “Core Computing Services” you will be presented with a list of topics that should direct you to the right type of request. Most of the requests are under the “Scientific Computing Services”, with a relevant exception in the creation of computing accounts, which is in “Core Computing Services”.

This ticket is completely unrelated to issue tracking tickets, opened via Redmine, against LArSoft or sbndcode pertaining LArSoft or ICARUS software bug reports or feature requests.

Accessing resources (DocDB, VOMS, …) via certificates

Note: this section is about “personal” certificates, not the kind of certificate proxies you need to work on the grid.

A CILogon certificate gives access to a number of Fermilab resources. After you get a CILogon certificate, you will gain access to:

Accessing resources via Virtual Private Network

Some resources can be accessed only from within Fermilab network. For some, the access outside Fermilab network is still allowed but crippled. To allow users to access these resources in full by connecting via a Virtual Private Network (VPN), which needs to be set up on the client node.

Submitting jobs: Virtual Organisation

The name of our virtual organisation is sbnd.

Submitting jobs: grid resources

This is about where your job gets actually shipped for execution, and who runs it.

We can use different resources:

  • FermiGrid is the grid dedicated to Fermilab users
  • Open Science Grid is an international grid, with strong contributions from physics but not limited to it
  • Wilson Cluster is a Fermilab cluster offering Graphics Processing Units (GPU)

To submit jobs, you’ll also need a certificate proxy.

Note: to ensure that your LArSoft jobs are executed in a complete environment, it is strongly recommended that the jobs are executed in the proper Singularity container (FIFE documentation here). One such container featuring Scientific Linux Fermi 7 (SL7) is available in CVMFS, and it can be enabled by adding to jobsub_submit the options: --append_condor_requirements='(TARGET.HAS_SINGULARITY=?=true)' -l '+SingularityImage="/cvmfs/singularity.opensciencegrid.org/fermilab/fnal-wn-sl7:latest"' (when using LArBatch project.py from larbatch v01_55_01 or newer, Singularity containers are already enabled in most configurations, as documented by the author, and often no action is needed, beside maybe adding a <os> element for good measure). Also note that FIFE recommends that Operating System selection be not enforced via --os option when using containers (after all, we are explicitly shipping the whole OS).

And, about FIFE recomendations, Project-py is the recommended job submission tool (SBN-specific wiki documentation pending).

Software distribution

SBND code and the “software stack” it relies on are distributed in binary form for a few supported platforms.

Here is where you can find those packages pre-installed: CVMFS and FermiApp. LArSoft/SBND code UPS products precompiled for selected platforms can also be downloaded from the SciSoft repository. The content of the areas is described in its own web page.

CVMFS

SBND CVMFS area is /cvmfs/sbnd.opensciencegrid.org.

CVMFS is a file system that mirrors locally a remote storage area (not dissimilarly from NFS), and it is available on the Open Science Grid nodes (and Fermigrid as well). So it is a safe bet.

You can also install it on your laptop; LArSoft links to instructions (currently from MicroBooNE). Good news is, it’s not hard.

The full path to access the area is: /cvmfs/sbnd.opensciencegrid.org/products/sbnd. The products/sbnd directory contains our software distribution.

[Instructions on how to deploy software in there are elsewhere.]

Storing data

We have two types of data storage available, that we’ll call “BlueArc” and “dCache”. FIFE explains their characteristics in detail. Production level data are stored to tape (“Enstore”) and tracked with a database (“SAM”).

Local storage: CephFS disks

CephFS is the filesystem used by Fermilab disk servers. These are multi-terabyte disks that are split between experiments. SBND has its own slice of it, in two partitions:

  • /exp/sbnd/data where you can write a few large files (e.g. ROOT data); if your data is larger than a hundred GB, people start muttering, as the space is limited and shared among all of us. Time to go dCache then.
  • /exp/sbnd/app where you can keep a few software builds; this is also a shared disk, and you should keep no large data here. But you need to have the code here, as content in the /exp/sbnd/data partition can’t be executed.

These data areas can be only directly accessed from the GPVM and BUILD nodes. If your grid job needs data from them, you have to copy themto the grid worker node first.

You should create your own directory under /exp/sbnd/data/users/${USER} and /exp/sbnd/app/users/${USER}, and stick to them. Space available to SBND collaboration:

path space
/exp/sbnd/app 6.0 TB
/exp/sbnd/data 25 TB

Be fair about disk usage and make sure to clean up unneeded stuff, this resource is shared by everyone in the collaboration.

The available disk space can be read using df -h.
Quota and usage per user are available in the SBND CephFS usage

World-visible storage: dCache

dCache area consists of a huge data space, made of tapes and disks.

We have three types of dCache areas:

  • scratch: files in this area can be deleted automatically; you can store a grid job result there waiting to transfer it into a safe place, but don’t rely on leaving your important stuff there for long
  • persistent: files in this area will not be deleted
  • resilient: area used for files with intense access (like a working area snaphot which is accessed by thousands of jobs)

We have a limited quota of space as well, so fill it responsibly. More information (including how to remotely access it) can be found in the SBND dCache wiki page.

Tape storage

20170906: SBND is requesting the ability to store data on tapes. This is a multi-step procedure requiring access to file transfer service, a file database and actual tape allocation.

File database (SAM)

Fermilab provides a service, SAM, to track where files are currently stored. An introduction to the system can be found on User Guide for SAM, and users interact with it via sam-web-client. In short, SAM not only stores metadata of the files (e.g., that a file is simulation, and how many events are in it) and the current locations (that is, the “original” tape location, the copies cached in dCache, the copies spread across worker nodes…), but it can also queue the files for delivery.

SBND is assigned a “station” that SAM clients will communicate with to access the database. More information on SBND SAM is collected in its own web page.

Web server pages

SBND and SBN has a number of web pages. A few are listed here.

SBN online web gate

The SBN online web gate is at https://sbn-online.fnal.gov. That is not a public page. Its content is also readable directly from SBND GPVM's and FNALU as /web/sites/s/sbn-online.fnal.gov. Direct write access is restricted, and can be requested asking William Badgett. Currently the site has almost no content and we are looking for volunteers to help populate it with useful detector status info.

SBND data web directory

The SBND data web directory is at https://sbnd-data.fnal.gov. The site is currently only accessible via a browser when connected to the FNAL network (VPN included). We are in the process of changing this. Its content is also readable directly from SBND GPVM's and FNALU as /web/sites/s/sbnd-data.fnal.gov. Direct write access is restricted, and can be requested asking the “owner” of the site. At time of writing, the site owners are Andrzej Szelc and Dominic Brailsford.

Communication tools

Mailing lists

Add link to mailing list webpage

Electronic logbook

SBND online system uses an electronic logbook (friendly called “eLog”), accessible at http://dbweb0.fnal.gov/ECL/sbnd/ . The Fermilab service account credentials are used to log in. Most current SBND members have an account. New members need to request one by clicking on the “Members” tab and then on the “Request an account” link.

Other obscure services and resources

SBND uses more services, which are usually hidden to most users. They are documented in their own page.