How to use MDSS tape storage at NCI

Revision as of 22:16, 3 August 2021 by P.petrelli (talk | contribs)

All universities have their own archives where you can store your data, for more information contact your Library or IT department. Here we will focus on archiving data using the NCI archive storage, for all cases where using a university archive is not applicable.

Massdata (Mass Data Storage System, MDSS for short) is the tape storage available at NCI. This kind of storage is intended for long term archiving of large files. Each project has a directory on the MDSS, the amount of storage allocated depends on the project allocation and can be checked using the nci_account command.

MDSS proper usage

MDSS is designed for medium to long-term archive of large files, this means it is capable and optimise for of storing big amounts of data, but not to retrieve this files often. This means it is most suitable for:

  • Files you are required to keep for a long term, like data underlining published datasets, publications, PhD thesis etc.
  • Files that you or someone else are likely to reuse or analyse again in the future but not in the next few months. For example restart files or other model output you are not immediately using should be moved from disk to massdata as soon as possible.
  • MDSS is suitable for backup of big data projects, like model output which could not be backed up elsewhere.


Preparing your data for mdss

  1. Organise your files and delete anything which you will not be re-using. Do not transfer data before organising it. It is difficult get a list of what is stored on massdata, let alone to list what is in a tarred file once it is uploaded. 
  2. Big files: use tools like tar to bundle files together into archive files. Create reasonably big archive files but also think of how you might want to access the data later. No point of tarring together two different simulations if you would want to access them separately, as then you would need to transferred back a big amount of data you do not need. Your upload will fail if any of your files are less than 20MB or the average size is less than 250 MB.
  3. Files should be group readable, with group execute permissions for directories. This helps with long term maintenance, allowing administrators to track the type and size of archived data.
  4. In the past you could use nci_account to monitor the allocation and how much of it was still available. Currently this is not possible anymore so, particularly if you want to move a big amount of data to massdata, you should first check with the lead CI of the project you want to use, to make sure enough storage is available.

While you are preparing your data to be moved it is an opportunity to also document, if you have not done so already, what you are archiving and how. Even a simple readme file added to your main directory can help others and your future self. If you are archiving data underlying a publication or published dataset then it is important to have a summary of what is stored in /massdata and how is part of the dataset management plan and/or data availiability statement.

Useful tools:

         TAR- to create archives cheatsheet

         Compressing tools

Accessing MDSS

Massdata cannot be accessed directly via a directory path. All access of MDSS is via the command mdss.

Users connected to the project have rwx permissions in the corresponding directory and so may create their own files in it.

Mdss has several sub-commands and options to see all of them:

mdss --help or
man mdss;

Usually you specify the project, if you don't it will use your default project, and then add a sub-command and the path of the files and directories you want to upload, list etc.

mdss -P <project-id>;+ <sub-command> + <path>

Most useful sub-commands are:

mdss put   - upload files 
mdss get   - retrieve files 
mdss ls    - list directories and files 
mdss dmdu  - get the size of a directory/file 
mdss dmls  - show what is on cache and what is on tape

NB "mdss du" will also work but only return the size of what is still cached, dmdu will give the full size of what is on tape regardless that is cached or not.

Please note mdss commands work only interactively or with ‘copyq’. To use it with copyq remember to set the storage flag as

      -l storage=massdata/<project_code>

Monitoring MDSS usage

Unfortunately, there is also not a command to check quickly usage by user-id as for /g/data and /short. The only way to get this information currently is to ask help<at>, administrators can access this information for any CI of the group.

Transferring data to and from MDSS

NCI also supports the netmv and netcp commands to work with MDSS. These commands create a copyq job to transfer multiple files. Files can be automatically tarred and compressed as part of the copy process.

NOTE: The process of compressing can use a lot of storage on /short if you're moving lots of data!

For more info run `man netmv`or check their User Guide.

The CMS team has also developed a utility called mdssdiff available from our conda environments. This utility allows users to compare the contents of the local directory and a directory under /massdata. It will also recursively update the content on the massdata directory to copy the local directory or vice versa.

Modifications to MDSS datasets

Contact NCI at help<at> if large metadata operations are needed on massdata, as changing ownership, project code, permissions etc. of existing datasets