Connection to interactive servers
CC-IN2P3 provides any user with a computing account two interactive clusters to prepare their tasks before they are submitted to the computing platform: development, test, etc. The general-purpose cluster cca
and the ccahm
cluster providing a large RAM capability and a local partition on SSD disk /scratch_ssd
.
Interactive clusters are accessible with the following syntax:
% ssh -Y <username>@cca.in2p3.fr
or
% ssh -Y <username>@ccahm.in2p3.fr
cca.in2p3.fr
andccahm.in2p3.fr
are aliases redirecting to the least loaded server of each cluster.For Microsoft Windows users we recommend the use of PuTTY as an
SSH
client.
For more information on your connection to the CC-IN2P3 servers please refer to the following pages:
Usage best practices
As each interactive server is shared by multiple users, we recommend a number of good practices to follow so that your activities can coexist with the other users activities.
Attention
Non compliance with these rules may result in some of your processes termination.
Only one computing process is allowed per interactive server and 3 processes over the entire interactive platform. Time restrictions apply:
24 hours maximum CPU time for a process and 1 hour CPU time for a multi-core process,
inactive processes for more than 24 hours are killed.
parallel processes with intensive use of both CPU and input/output are not allowed (example:
make -j
,scons
,b2
,ninja
, etc.). However, it is possible to run such processes as interactive jobs (see interactive jobs),avoid excessive memory consumption that could have an impact on the system, if large memory usage is required, please switch to the ccahm cluster . The command
htop -u username
may be a useful monitoring tool,it is forbidden to run tasks in “batch” mode (eg: processing a large amount of files in loop),
/scratch
is not a permanent storage space: data can be deleted at any time to free up space. Under normal circumstances data is deleted after 14 days. Also, avoid large data storage,One transfer process is allowed per server (example:
scp
,rsync
ordq2-get
) and 3 processes across all interactive servers,Detached and persistent processes after logout are not allowed.
Screen may be an exception to this rule, but it is neither meant as a substitute to the batch system, or as a way to run services nor as a way to hog scarse resources (like GPU interactive computing nodes),
the usage case of Visual Studio Code is explained in the dedicated paragraph.
To preserve user data confidentiality,
rmate
(or other similar editor) requires using a random TCP port instead of a fixed port.
Data transfer
Only the SSH protocol is allowed for connections and file transfers to or from the CC-IN3P3 interactive servers. Two authentication modes are preferred: using password, or Kerberos ticket (via GSS-API). An RSA/DSA type authentication (SSH key) will grant you access, but will not initialize your Kerberos ticket.
To copy files:
% scp <my file> <username>@cca.in2p3.fr:~/<my directory>/
% sftp <username>@cca.in2p3.fr
For more information, please refer to the above commands manual pages (man <command>
).
Change shell
To change your default shell, use the command changesh
with the following syntax:
% changesh --help
Usage: changesh --help | -h
changesh <shell>
Where <shell>
could be bash
, csh
, ksh
, tcsh
or zsh
.
Usage of Visual Studio Code
If the internet connection on your workstation may be interrupted by your screen standby (the case of Mac users with a Thunderbolt connexion, i.e.: a connection to the Ethernet network wired on the screen), your VS Code windows will be closed due to the lack of connection. Below is the description of what is happening:
VS Code creates an SSH connection to a
cca
server and launches (on the interactive)Node.js
,the user temporarily interrupts his activity,
the workstation detects the lack of activity and puts the screen on standby: the network is disconnected and
the SSH connection disappears,
but VS Code’s
Node.js
persists,the ghost VS Code is detected by the interactives monitoring (and killed).
On the other hand, and by default upon startup, VS Code looks for the names of the symbols in the source code for its auto-completion functions. For this purpose, it uses ripgrep
which uses all the cores of the machine to do a quick scan.
It is recommended, if possible, to disable this collection with ripgrep
(or to configure it to be less aggressive) to avoid too much load on the interactive server.