Using Rotterdam
`Rotterdam' or `rdam' is the name of the server financed by the Department of Econometrics & O.R., or NWO, at the VU University Amsterdam. Members of the department can get an account from Laurent Callot or Charles Bos. The server is meant for all kind of computational jobs, as long as they can run on Linux.Some definitions
- `Rdam' is the machine for which you got an account.
- It is known by IP address 145.108.238.26. It is reachable from outside the university as well, at this address.
- You received your own login and password. Your login is personal, you cannot give your password to anybody else.
- In the text below, I'll use VUNetID (abc123) as example. Replace it with your login where needed.
- The present machine has 2 processors of 8 cores each, and is capable of hyperthreading on each core. Hence, you could run up to 32 processes to use the full machine. There is 32GB of memory, and plenty of local hard disk space. Furthermore, the machine contains a Tesla card, for GPGPU computations.
- Each user has a limited amount of priority, e.g. if two users are working at the machine usually each is limited to 50% of the cpu-time, if there is a conflict. For the memory, there is no automatic limit. Please make sure you do not use too much memory, else the machine becomes very slow. On the other hand, 32GB seems to be enough for plenty of programs.
Entering the machine
You can enter the machine using a secure shell program, e.g. PuTTY. At the VU, send an email to ucit-servicedesk@vu.nl that you want access to PuTTY (if you don't have it already). At home, download and install http://the.earth.li/~sgtatham/putty/latest/x86/putty.zip.Start PuTTY, fill in as host name
145.108.238.26and save the session under the name `Rdam'. Then hit the Open button.
Fill in your login and password. If this is the first time you login, you will be asked to change your password, do this using
passwdYou would do well to change it to your VU password.
When you finish with PuTTY, type
exitor
ctrl-dor just kill the window.
For just getting to the files (from home) you could use WinSCP or FileZilla. This gives you a graphical environment to copy files to and from Rdam. FileZilla is also available on the VU.
Using your own files
After the migration (and if I have set up things straight for you, complain in case of errors), you can usemount ~/hdriveto mount your h: drive. The command may ask for your username (if I didn't set it for you already). This would be
vu/abc123(ie vu/ followed by your vu-net id, don't forget the trailing vu/!). The password is the standard password you have on the VU network.
Lately this link to the network drive seems to work. Only with many large files at once I have some problems at times. If all else fails, work in a local directory.
With a
cd hdrive/oxprogsyou would get into the
h:\oxprogsdirectory (assuming you have one, of course). Note the forward slashes in directory names.
If you would wish to unmount the drive, you could use
cd umount ~/hdrive
Using local files
You could use local data files (which might go slightly quicker if the files are very large) or save your output locally. All files which are not saved underneath the /home/abc123/hdrive directory will end up locally. For the moment, feel free to use a couple of GBs, please ask if you need more.Running commands
On the machine you can use Ox, gcc, octave, matlab and possibly other programs as well. Run e.g. Ox usingcd cd hdrive/oxprogs oxl myprogif all packages would be available. Ox on Rdam has the standard packages around, but you can add extra ones by creating
h:\bin\ox\packages\oxprobfor instance (the $HOME/bin/ox/ directory is, on Rdam, automatically added to your OX7PATH variable). If you place the oxprob package there, creating the necessary directories, Ox should be able to find the files. Note that also Ox on your Windows machine at the VU will search this same directory, hence also on your desktop you should be able to use this same package.
Command control
You can run a job in the background, by typingoxl myprog &This way, you could even run multiple jobs at once. If you have a job running, you can leave PuTTY using
ctrl-d(if you leave PuTTY any other way, your job will stop).
When you login to Rdam anew, you can use the command
ps xto see a list of jobs you have running, or use
topto see the most active jobs of yourself and others. Stop the `top' with ctrl-c.
To run multiple jobs on Rdam, you can start working with bash-scripts. To track the number of jobs you have running, you could use something like
for i in `seq 1 100`; do oxl mysim seed $i & checkcpuload done waitwhich would start 100 iterations of mysim, each with a different command line argument for the seed. The checkcpuload will could the number of Ox jobs you have running, and wait until a core is free. At the end, the wait command will wait until all jobs are finished.
Here I use the local command checkcpuload. With a
checkcpuload testyou get some output on screen on the number of Ox jobs you seem to have running already. Note that checkcpuload is a local command, not available by default on sara/lisa.
Ox and multitasking
Contrary to standard installation (as on Sara for instance), here on Rdam the default is to run Ox on one thread at a time. That is to say, by default automatically the option -rp1 is added.In case you know that running multithreaded Ox makes sense for you, use either
oxl -rp4 mymultithreaded.oxto test it out, or set it with a
export OXARGS="-rp4"to run using 4 threads. You could adapt
export MAXCNT=4to tell checkcpuload to run a maximum of 4 Ox jobs now, instead of the 32 it usually allows for.
On Rdam, you use a beta version of Ox7, including the option to use parallel for. Please read chapter 4 of the manual, available from Rdam at /usr/share/OxMetrics7/ox/doc/Ox.pdf.
Dropbox
Dropbox functionality is available, if- you inform me that you'd like to use it
- after I start the service, use a command
dropbox start -i
and follow the resulting link using the lynx browser from the Rotterdam machine, to fill in your credentials.
Matlab and multitasking
Also Matlab is available on the machine. Within Putty you could startmatlabdirectly, and work from there. This has the drawback that you have to leave PuTTY running until your program finished. Alternatively, start your scripts using
m myscript_1.m & m myscript_2.m & m myscript_3.m &which runs your script in the background. You can now log out using
ctrl-d(NOT with exit), leaving the scripts running.
Other commands
Maybe check some Linux introduction on the manual forcd | Change directory |
top | See what jobs are running |
fg | Get a stopped or background jobs in the foreground |
bg | Put a stopped job (after pressing ctrl-z) in the background |
ls | Show directory contents |
mkdir mydir | Make a new directory with name mydir |
more x.txt | Show contents of x.txt |
cp a b | Copy file a to b |
rm a | Remove file a |
passwd | Change password |