The HaxbyLab@Dartmouth

Software

«  Hardware   ::   Home   ::   Network  »

Software

Initial Software Adjustments

  • installed some basic stuff:

    apt-get install etckeeper git-core sysstat iftop htop sysv-rc-conf \
     sysv-rc-conf sun-java6-plugin
    

    sun-java6-plugin was installed to enable remote console within iceweasel through IPMI

    On main node (head1):

    logwatch dlocate dpkg-dev
    
  • backported/uploaded/installed cfengine3 . Basic configuration is in place. To cause updates of configurations across nodes use:

    cf-runagent -o '-b update'
    

    Configuration would also be tried to be updated automagically if current one available at a node is faulty.

  • installed nut (lenny’s version)

    Configuration
    • copied/adjusted templates into /etc/nut
    • adduser nut dialout # to access serial
    TODO
    configure/enable upsmon:
    • copy/rename upsmon.conf.sample upssched-cmd upssched.conf.sample from /usr/share/doc/nut/examples/
  • domain name change: although Microway advised not to do so, on head1 I have done:

    cd /etc
    sed -i -e 's/cl\.dartmouth.edu/hydra.dartmouth.edu/g' \
          dhcp3/dhcpd.conf lynx-cur/local.cfg  hosts defaultdomain printcap resolv.conf*
    
  • reconfigured exim4 so it could serve us mail

  • update-alternatives --config editor to point to vim

  • disable torque/maui tandem:

    sysv-rc-conf pbs off
    

    They didn’t bother to package torque/maui so it is installed under usr/local, so might be tricky to get completely away from this planet.

  • custom banner /etc/issue.net

  • Packages removed from all the nodes

    avahi-daemon gnome libnss-mdns

  • Packages removed from computing nodes

Backup

We are using BackupPC to backup head1 and possible some other things. head2 is the backup RAID and that is where backuppc was setup to run. It access head1 node via rsh and calls /usr/local/sbin/tarCreate wrapper.

Logging

rsyslogd on head1 allowed

$ModLoad imudp

Monitoring

Access Ganglia status page when you are at Hydra.

TODO

  • remove mdadm from the nodes – no need

Software not coming from Debian proper

There is quite a bit, but some of it we would not care.

We don’t care about

  • torque/maui
  • ...

But we should care about

  • Available under /usr/local/bin:

    anytermd

    Should allow serial console from WEB GUI (may be we shouldn’t care whenever we setup proper SOL way)

  • Available under /usr/local/sbin

    ipmimon

    Collects and makes available health status from the nodes.

    TODO: it seems to query more nodes/heads than we have

  • Available under /usr/sbin

    3dm2

    Interface to 3ware card. Installed not via package, but binary packages available from http://debian-unofficial.org (TODO)

IPMI

Setting up a new user

Using ipmitool can carry out control over IPMI module on MB:

>sudo ipmitool -I open shell
ipmitool> user set name 3 yoh
ipmitool> user set password 3 XXX
ipmitool> user enable 3
ipmitool> user priv 3 4

Load IPMI modules whenever:

modprobe ipmi_devintf
modprobe ipmi_si

TODO (so far was just done partially in web interface for head1)

  • Use NTP for time synchronization (might need zone specification)
  • Enforce https
  • Enable ssh access (though not yet sure what to do with it)
  • Enable SMTP logging
  • Figure out MAC addresses? ;)
  • Configure mouse so they stay in sync

EDAC

For now just manually added loading of e752x_edac module to /etc/modules.

Bootstrapping of new system

cd /etc/apt; rcp head1:$PWD/{sources.list,trust*.gpg} .; apt-get update; apt-get install cfengine3
cd /var/lib/cfengine3/ppkeys/; rcp localhost.pub head1:/var/lib/cfengine3/ppkeys/root-`ifconfig bond0 | sed -n '/inet addr/s/.*addr:\(.*\)  Bc.*/\1/p'`.pub; rcp head1:/var/lib/cfengine3/ppkeys/localhost.pub /var/lib/cfengine3/ppkeys/root-10.0.0.1.pub
cd /var/lib/cfengine3; mv inputs masterfiles; mkdir inputs; cd inputs; rcp head1:/etc/cfengine3/{update.cf,failsafe.cf} .

«  Hardware   ::   Home   ::   Network  »