The only way to access SURFsara's HPC systems is by using ssh, scp or sftp.
It is commonly known, that a normal telnet or ftp connection to a remote system can be very unsafe. In many cases it is possible to use a computer to catch all tcp packets that are put on the local network. All data that is send via the local network to a remote (or local) computer system can be read by anyone who has access to a computer that is linked to the local network. This is especially very undesirable in the case of passwords that are sent to other systems. The SSH standard circumvents this problem, by encrypting all data that is sent over the network, also the passwords needed for authentication.
How: installation on your work station
Installation of the software involved is, as far as we experienced, rather simple. The central site for the open SSH is http://www.openssh.com.There one finds implementations for Unix-es, Windows, Macintosh. In the following we will elaborate somewhat on the different platforms:
/usr/local/ssl/lib/or other "standard" locations. Most, if not all, linux distributions support SSH. Installation from source is easy, one doesn't even need root permission to install the ssh and scp (replacements for telnet and ftp) in one's home directory.
ssh -X email@example.com or ssh -X firstname.lastname@example.org
where myname is the login name on the HPC system, and hpc-system.surfsara.nl is the internet address of the HPC system. The -X flag is for the automatic arranging of your X environment. It depends on the configuration of ssh on your local workstation if this flag is necessary.
scp infile email@example.com:dir/tofile
infile is the file you want to copy (you can specify more than one file and use wildcharts if you like) and
directory is the destination directory on the HPC system. When you omit
directory, the file(s) will be copied to your home directory.
tofile is the name the file gets on the HPC system. When omitted, the name will be the same as infile.
scp firstname.lastname@example.org:dir/infile outfile
infile is the name of the file on the HPC system,
outfile is the name of the file on your workstation.
There is a famous security problem connected with using X11 applications if your workstation is running some kind of Unix. Many people used the xhost command to produce X11 output on their screens. Not everbody is aware of the fact, however, that using the xhost command one creates very likely a situation where somebody, also logged in on the remote system can get access to the X-server running on the workstation and is able, for example, to read out every keystroke.
So we cannot stress enough: DON'T USE XHOST.
An alternative and more approved approach is the use of 'cookies' with the 'xauth' command.
Using ssh, neither of xhost or xauth is necessary, as ssh takes care of the X-window traffic, and setting the DISPLAY environment variable. Once you have a ssh connection from your Unix workstation with a SURFsara system, X-applications will automatically put their windows on your workstation, and all traffic is encrypted. Depending on the configuration of your system, you have to add the -X flag to the ssh command. When using PuTTY under Windows, choose 'Tunnels' from the 'Category' list and check the option 'Enable X11 forwarding'. Please note that you need a separate X-server, e.g. http://www.straightrunning.com/XmingNotes/.
Another bonus is the possibility to login without using a password. For details, have a look at http://www.openssh.com, but in its most simple form you could do the following:
ssh-keygen -t rsa
id_rsa.pubis located there. Append this file to the file
$HOME/.ssh/authorized_keyson the HPC system. You can log in from your workstation to the HPC system using the pass-phrase.
ssh-addafter which you will be prompted for your pass-phrase. The system remembers your pass-phrase and it is supplied automatically with subsequent ssh and scp commands.
and give the
If the directory $HOME/.ssh does not exist on the HPC system, you can create that directory:
mkdir -m 0700 $HOME/.ssh
When you make a SSH connection to a remote system for the first time, SSH logs in the file
$HOME/.ssh/known_hosts the hostkey of that system and asks if you want to proceed. If you enter yes, the hostkey of the remote system is stored in
$HOME/.ssh/known_hosts on your local system. When you make later on again a SSH connection to the same system, SSH compares the hostkey with the stored hostkey. If there is a difference, you will get a warning and some functionality is not available. This is to protect you from possible attacks: you think you are connected to a particular host, but in reality you could be connected to another host (the so called "Man in the middle attack").
Most of the times, however, there is no problem because the hostkey of the host system has been changed for some reason, perhaps by upleveling the system. If you are sure that you are really connecting to the system you want, simply remove the line with the name of the remote host in the file
$HOME/.ssh/known_hosts and connect again. You could also try a more drastic approach: remove the file
$HOME/.ssh/known_hosts altogether. The above is for Unix systems. In general, the SSH implementations for Windows and Macintosh issue a warning and let you choose to change the stored hostkey.
For your verification, the RSA key fingerprints of our systems are listed here.
|system||host||RSA key fingerprint|
SURFsara considers easy and secure access to the HPC systems important, so in case of difficulties, please contact email@example.com.