Accessing Files

From Molecular Modeling Wiki

(Difference between revisions)
Jump to: navigation, search
(Notes)
 
(43 intermediate revisions not shown)
Line 1: Line 1:
-
= Problem =
+
== Problem ==
-
Data in the cluster home directories usually need some processing after they are produced. Frequently, some visualization of data (inputs, results) is required and programs like '''molden'', '''VMD''', and others are often run on the cluster servers to display images on the user's workstation through the X-Window system. Howvere, this method has disadvantages - it is slow, as a huge amount of graphical data is transferred through the network (especially when the object consisting of many atoms moves and rotates), and it sometimes does not waork at all (mostly because of incompatible versions of OpenGL libraries on the cluster server and user's workstation). There is a better solution, though.
+
Frequently, some visualization of data (inputs, results) in user's $HOME directory is required and programs like '''molden''', '''VMD''', and others are often run on cluster servers to display images on the user's workstation through the X-Window system. However, this method has '''major disadvantages''' - it is slow, as a huge amount of graphical data is transferred through the network (especially when the object consisting of many atoms moves and rotates), and it often does not work at all (mostly because of incompatible versions of OpenGL libraries on the cluster server and user's workstation). The solution is to run visualization programs locally (on user's workstation) and bring data to the program, instead of bringing the program to data. For this, you do not need to '''copy''' data from server to your workstation - there is a way to '''embed''' (''mount'') your remote $HOME directory to your local file system.
-
= Idea =
+
== Sshfs ==
-
''If the mountain won't come to Muhammad then Muhammad must go to the mountain.'' Or translated to our case: If you cannot run the graphical program on the remote server to let it work with the local data, make the data local on your workstation and run the program there. In other words - install visualization programs on your workstation where they will be fast and configured exactly for your system and make the data appear as if it is local too. For this, you do not need to copy data from server to your workstation - it is enough just to embed your remote home directory to your local file system. Read on for instructions.
+
One of the systems which allow to mount the remote directory to local filesystem is called ''sshfs''. To use it, follow these instructions (specific for Ubuntu, see also this [https://help.ubuntu.com/community/SSHFS howto]):
-
= Sshfs =
+
Install the sshfs package and add yourself to the ''fuse'' group (this has to be done only once)
-
One of the systems which allow to '''mount''' the remote directory to local filesystem is called ''sshfs''. On Ubuntu, you can follow these instructions:
+
<pre>
 +
sudo apt-get install sshfs
 +
sudo gpasswd -a $USER fuse
 +
</pre>
-
* First, add yourself to the ''fuse'' group
+
Once you have added yourself to the ''fuse'' group, you should log out and log back in again for the change to take effect.  
-
<pre>sudo gpasswd -a $USER fuse</pre>.
+
-
This has to be done only once. Once you have added yourself to the fuse group, you should log out and log back in again for the change to take effect.  
+
-
* For every filesystem you want to ''mount'' or ''connect'', do
+
For every filesystem you want to mount do
-
<pre>cd ~
+
<pre>
-
mkdir -p mnt/_SERVERNAME_
+
mkdir -p ~/mnt/SERVERNAME
-
sshfs -o idmap=user _USERNAME_@_SERVERNAME_:/home/_USERNAME_ ~/mnt/_SERVERNAME_</pre>
+
sshfs -o idmap=user USERNAME@SERVERNAME:/home/USERNAME ~/mnt/SERVERNAME
 +
</pre>
-
replacing _SERVERNAME_ with the name of remote server (cluster) and _USERNAME_ with you login name on this server (cluster) for example
+
replacing SERVERNAME with the name of remote server (cluster) and USERNAME with your login name on this server (cluster). For example
-
<pre>cd ~
+
<pre>
-
mkdir -p mnt/zinc
+
mkdir -p ~/mnt/zinc
-
sshfs -o idmap=user novak@zinc:/home/novak ~/mnt/zinc</pre>
+
sshfs -o idmap=user novak@zinc:/home/novak ~/mnt/zinc
 +
</pre>
 +
 
 +
{{Note|This process can be repeated for as many cluster servers as needed.}}
 +
{{Note|The directories (mountpoints) need to be created only once for every remote directory.}}
 +
 
 +
Once the remote directory is mounted, you can use it as if it was local, so, for example, you can call your local installation of '''VMD''' or '''molden''' and open a (remote) file in it browsing to ''~/mnt/SERVERNAME/...''
To unmount (disconnect) the remote directory from your local filesystem call
To unmount (disconnect) the remote directory from your local filesystem call
-
<pre>fusermount -u ~/mnt/_SERVERNAME_</pre>.
+
<pre>fusermount -u ~/mnt/SERVERNAME</pre>
-
Once the remote directory is mounted, you can use it as if it was local, so, for example, you can call your local installation of VMD and open a remote file in it browsing to ~/mnt/zinc/...
+
== Notes ==
 +
This method can also be used to mount cluster $HOME directory to your home computer. For this, however, opening the firewall with [[Access_from_the_Internet#Port_knocking_method|knock]] is needed before calling ''sshfs''.
-
<pre>
+
The same ''sshfs'' functionality in Windows XP/7/8 can be achieved by open source software called [http://dokan-dev.net/en/download/ Dokan SSHFS] (credits to Lukasz Cwiklik).
 +
 
 +
[[Category:Clusters]]

Latest revision as of 10:44, 29 January 2014

Problem

Frequently, some visualization of data (inputs, results) in user's $HOME directory is required and programs like molden, VMD, and others are often run on cluster servers to display images on the user's workstation through the X-Window system. However, this method has major disadvantages - it is slow, as a huge amount of graphical data is transferred through the network (especially when the object consisting of many atoms moves and rotates), and it often does not work at all (mostly because of incompatible versions of OpenGL libraries on the cluster server and user's workstation). The solution is to run visualization programs locally (on user's workstation) and bring data to the program, instead of bringing the program to data. For this, you do not need to copy data from server to your workstation - there is a way to embed (mount) your remote $HOME directory to your local file system.

Sshfs

One of the systems which allow to mount the remote directory to local filesystem is called sshfs. To use it, follow these instructions (specific for Ubuntu, see also this howto):

Install the sshfs package and add yourself to the fuse group (this has to be done only once)

sudo apt-get install sshfs
sudo gpasswd -a $USER fuse

Once you have added yourself to the fuse group, you should log out and log back in again for the change to take effect.

For every filesystem you want to mount do

mkdir -p ~/mnt/SERVERNAME
sshfs -o idmap=user USERNAME@SERVERNAME:/home/USERNAME ~/mnt/SERVERNAME

replacing SERVERNAME with the name of remote server (cluster) and USERNAME with your login name on this server (cluster). For example

mkdir -p ~/mnt/zinc
sshfs -o idmap=user novak@zinc:/home/novak ~/mnt/zinc
Note Note: This process can be repeated for as many cluster servers as needed.
Note Note: The directories (mountpoints) need to be created only once for every remote directory.

Once the remote directory is mounted, you can use it as if it was local, so, for example, you can call your local installation of VMD or molden and open a (remote) file in it browsing to ~/mnt/SERVERNAME/...

To unmount (disconnect) the remote directory from your local filesystem call

fusermount -u ~/mnt/SERVERNAME

Notes

This method can also be used to mount cluster $HOME directory to your home computer. For this, however, opening the firewall with knock is needed before calling sshfs.

The same sshfs functionality in Windows XP/7/8 can be achieved by open source software called Dokan SSHFS (credits to Lukasz Cwiklik).

Personal tools