How to Transfer Data

From DeepSense Docs
(Redirected from Transferring Data)
Jump to: navigation, search

There are different methods for transferring data to and from the DeepSense platform. Which method you use will depend from where you are transferring the data, as well as the size of the data.

To and From Your Personal Computer

Small Transfers

For small transfers (<5Gb), you can use the two login nodes. Since they are the primary point of access for the platform, they may be in heavy use. We do not want to overload them unnecessarily for data transfer. Please only use this for small amounts of data.

The most common method for transferring data securely between machines will be scp. This is pretty straightforward to use, however the destination files will have the wrong permissions set. It will remove group permissions, so while you will be able to access the data, no one else in your group will be. This is fine if you are the only one working on the project.

Example: scp -r /path/to/files/

One can also use rsync (see the man page). This has more options than scp, and can be used to sync files between two machines.

Example: rsync -azvhP /path/to/files/

The rsync options above are:

  • a - archive mode, equal to rlptgoD (recursive, preserve links, times, permissions, group, owner, etc)
  • z - use compression when copying
  • v - verbose: list files copied
  • h - human readable: output numbers in human readable format
  • P - same as --partial --progress. Show progress while transferring, and keep partial files.

Note: We recommend always using the option -p (using -a also invokes -p). This ensures that everyone in your group should have the same permissions to the file as you do.

Medium Size

For medium sized transfers (between 5Gb and 100Gb), you should use the protocol nodes. They (, are specifically meant for data transfers. However, they are only accessible via samba.


Connect via samba on OSX

On a Mac, open finder and hit ⌘-K, or use the menu Go -> Connect to Server. In the dialog box (see image), type the address for either protocol node, and you can login. This will connect you to the /data filesystem.

If you want to use rsync to transfer data via the protocol nodes, you have to mount one. On a Mac, the easiest way is to connect to the protocol node as in the previous paragraph. This will mount it at /Volumes/data/. You can now use rsync to copy files to your project's subdirectory.

Example: rsync ‑rzvh /path/to/files/ /Volumes/data/projectdir/


On windows computer, you should connect to \\\data or \\\data. To do this the first time, open a file explorer window. Right-click on This PC, and select "add a network location". In the wizard, click next and then select "Choose a custom network location" (this was the only option I saw). Highlight it, and click next. On the following screen, enter one of the addresses above, and click next. You may now enter a name for this location. Do so, and click next again. On the last screen, you should be able to look over your selections, and then click Finish. The name you chose should now be available under "This PC" in your file explorer.

You may also have to change a SMB security level setting as follows (this was necessary in Windows 10):

Control Panel > System and Security > Administrative tools > Local Security Policy > expand Local Policies > Security options > click on Network security: Lan Manager authentication level > Then in the field choose > Send NTLMv2 responses only > click on Apply, then ok and close all.

File Permissions

Unfortunately, samba won't preserve the proper file permissions. We find it strips the executable bit from any file that has it switched on. You can change an individual file by using chmod ug+x filename. If you want to change many files at once, and are unsure of how to, please send us an email at (

Large Transfers

For large transfers (>100Gb), We generally find it is best to put the data on an external drive. To make such arrangements, please email We can then plug it in directly in our server room, and transfer the data for you.

From the World Wide Web

The standard tool for downloading data from websites is wget. Also available is curl. The two are compared in this StackExchange article.

If you have URLs for multiple datasets, you can also use python code (or others) to download the files you need. You can write a script that will look like this:

import urllib
urls=[ "url1", "url2", ...]


for url in urls:
  urllib.request.urlretrieve( url, filename=destination)

Of course, you'll have to properly specify the filename destination.