[Vagnekman-users] Usage advisory of the login node on Vagn

[Johan Raber]

Dear Vagn Users,

During the recent meeting with your user representatives it came up that 
as of late (last month) the interactive response of the login node has 
felt sluggish to some users and also there has been a dissatisfying 
number of crashes due to its usage. For a few weeks now, the login node 
has been taken out of the batch queue which should ameliorate these 
issues to an extent. In this context, remember that you are sharing this 
resource and load will be high at times. However, please be aware that 
your usage of the login node affects many other users. Take a look at 
how many on a typical day by issuing the command "w". I will hate to 
have to start policing the login node usage.

The login node can be used, but responsibly and with measure. For 
instance Matlab is a fickle beast when it comes to consuming computer 
resources, please be very careful with its use on the login node or we 
will eventually be forced to consider drastic measures, such as not 
allowing access to it on the login node.

A few (impertinent, I know) advisories:
* If you note extraordinary sluggishness on the login node, or indeed 
any problem on any node, please report it immediately to 
vagnekman-support at snic.vr.se. Remember to mention specifics if you can 
such as node, time of day, job id and output from diagnostic commands 
(top, ps, vmstat etc.) not to mention which system (Vagn or Ekman) it 
concerns, we are not psychics.
* Use the "share" partition (as in "interactive -p share" for instance) 
unless you are absolutely sure you need an entire node's resources. In 
my unscientific sampling (~5) of noshare jobs I have only seen single 
processor jobs running. This is a serious waste in my opinion.

Best Regards,
Johan Raber -- NSC support