Data Transfer Tools: GridFTP (SSH authentication)

This article describes how to transfer data using GridFTP with SSH authentication. It covers:

  • Introduction
  • Establishing a connection
  • Example GridFTP usage
  • Upload data (push data from JASMIN to remote server)
  • JASMIN host as remote server

Introduction

GridFTP is the recommended tool for transferring large files or groups of files across high-speed Wide-Area Networks (WANs). It is commonly used with certificate-based authentication, but can also take place between suitably configured* server and client using SSH as the authentication mechanism.

*The client may need to have certain ports enabled on the host or institutional firewall if present. Consult your local IT support desk for details and direct them to the  client section of "3. Firewall requirements" in the GridFTP System Administrators Guide.

SSH-based GridFTP does not enable the full feature set provided by certificate-based GridFTP and in particular does not work with Globus Online, which provides useful interfaces and APIs for managing large-scale data transfers, but still provides a major step up in performance by "filling the pipe" more efficiently than scp/rsync/sftp, particularly over longer distances and can do verification and sync operations as part of the transfer.

See also:

  • Transfer Servers for details on which servers within JASMIN have GridFTP available.
  • Transfers to/from ARCHER/RDF for details of different routes affecting your choice of server (since this is the one of the most likely places to which JASMIN users will want to transfer data to/from)

Establishing a connection

Since you will be using SSH as the authentication mechanism, you should ensure that your initial connection to the JASIMN transfer server is made with the -A option enabled, to enable agent forwarding:

$ ssh -A username1@jasmin-xfer2.ceda.ac.uk

Note that in order to use jasmin-xfer2, you will need to have high-performance data transfer access on your JASMIN account. An alternative to try out beforehand, or if you intend to use the light path connection to RDF is to use jasmin-xfer1.ceda.ac.uk.

Use the globus-url-copy command to list the contents of your home directory on the remote server (This will only work if you already know that that server supports GridFTP over SSH). In this case, we are making the connection to one of the data transfer nodes at RDF, dtn02.rdf.ac.uk:

$ globus-url-copy -vb -list sshftp://username2@dtn02.rdf.ac.uk/

If username and username2 are the same (on the different systems, the username@ part of the sshftp URI can be omitted.

Note that the URI of the server, in this case sshftp://username2@dtn02.rdf.ac.uk must come immediately after (as it is an argument to) the -list option. This is particularly important if you are combining this command with other options.

Example GridFTP usage

Once you have successfully established that you can connect to the server (as above), then you should be able to transfer data between the remote end (server) and local end (client) with commands such as shown below:

As above, if you have the same username on both local and remote systems, then the  username@ part of the sshftp URI can be omitted.

Please consult the documentation for the  globus-url-copy command for the full range of options and arguments.

$ globus-url-copy -help

See also   http://toolkit.globus.org/toolkit/docs/latest-stable/gridftp/user/#gridftp-user-basic

2. Download a file from remote directory  /home/users/USERNAME to destination on the local (client) machine, for example a group workspace on JASMIN:

$ globus-url-copy -vb sshftp://username@gridftp.remotesite.ac.uk/home/users/USERNAME/myfile /group_workspaces/jasmin/myworkspace/myfile

The  -p N and -fast options can additionally be used in combination to enable N parallel streams at once, as shown below. You can experiment with N in the range 4 to 32 to obtain the best performance, but please be aware that many parallel transfers can draw heavily on shared resources and degrade performance for other users:

$ globus-url-copy -vb -p 16 -fast sshftp://username@gridftp.remotesite.ac.uk/home/users/USERNAME/myfile /group_workspaces/jasmin/myworkspace/myfile

3. Test performance with large files by downloading from /dev/zero on the remote server to /dev/null locally. This excludes any interaction with either filesystem and gives an upper limit to the performance that can be achieved at the time. Repeat with values of N in the range 4 to 32 to compare rates. Note that the performance takes a while to "ramp up", so you will not see the best rates if transferring small files individually as the process never gets up to full speed:

$ globus-url-copy -vb -p 16 -fast sshftp://username@gridftp.remotesite.ac.uk/dev/zero /dev/null

Press CTRL-C to interrupt the transfer. Alternatively you can specify that the transfer should continue for a fixed duration in seconds using the -t option. In this example, data is transferred from the RDF data transfer node dtn01.rdf.ac.uk to jasmin-xfer1.ceda.ac.uk.

[username@jasmin-xfer2 ~]$ globus-url-copy -p 16 -fast -t 10 -vb sshftp://username@dtn01.rdf.ac.uk/dev/zero /dev/null
Source: sshftp://username@dtn01.rdf.ac.uk/dev/
Dest:   file:///dev/
  zero  ->  null

   7797473280 bytes       929.52 MB/sec avg      1024.49 MB/sec inst
Cancelling copy...

Note the transfer rate achieved in Megabytes/second (MB/sec), although for various reasons this is not to be relied upon as an accurate expectation of speed for real transfers. However, you are unlikely to achieve even half of this data rate via scp, rsync or sftp over the same routeBbcp may achieve similar rates, however, and this is considered by some as easier to use.

4. Recursively download the contents of a directory on a remote location to a local destination.

$ globus-url-copy -vb -p 4 -fast -cc 4 -cd -r sshftp://username@gridftp.remotesite.ac.uk/home/users/USERNAME/mydir/ /group_workspaces/jasmin/myworkspace/mydir/

Where:

  • -cc N requests N concurrent transfers (in this case, each with p=4 parallel streams)
  • -cd requests creation of the destination directory if this does not already exist
  • -r denotes recursive transfer of directories
  • -sync and -sync-level options can be used to synchronise data between the two locations, where destination files do not exist or differ (by criteria that can be selected) from corresponding source files. See -help option for details.

Upload data (push data from JASMIN to remote server)

The above commands can also be adapted to invoke transfers from a local source to a remote destination, i.e. uploading data, since the commands all take the following general form:

$ globus-url-copy [OPTIONS] source-uri desination-uri

Be sure to check your connection with the remote machine via a simple SSH login and then a directory listing as shown above.

JASMIN host as remote server

So far the examples have used a server within JASMIN as the client in the GridFTP transfer. The transfer can be reversed so that the client is elsewhere and the JASMIN host is the server specified in the destination URI. The following command should work connecting to one of the following transfer servers: (see also  Transfer Servers)

Push data to JASMIN from a remote server:

$ globus-url-copy -vb -p 8 -fast mydir/myfile sshftp://username@jasmin-xfer2.ceda.ac.uk/group_workspaces/jasmin/myworkspace/mydir/

Still need help? Contact Us Contact Us