Operational Meeting

Alison, Craig, Gordon

These were accepted. Neil noted that there had actually still been two services-unit servers in FH at the time of the last meeting.

Comments:

• Neil's changes to the amd resources look OK. They can be tested by adding a #define to a machine's profile. mp-unit will be testing them.

  The issue of /home was raised. We might be able to break it on servers, which is where we intend removing amd in the first instance, but we can't break it on desktops. Might autofs provide an answer?

Actions completed:

• Add profile resources for machine criticality -- on mp-unit's list
• Alastair has asked about the Building Committee minutes. Apparently they haven't been published yet, to the BC's surprise.
• Self-managed machine shutdown -- discussed later
• Define criticality of machines -- now waiting for profile resources

Actions changed:

• mp-unit to take over the "cups on servers" action, and to consider a more general change

Actions added:

• Tim to inform Teaching Committee and/or Stuart about the impending demise of the news service
• mp-unit to document how to handle the power button
• inf-unit to set up a front page for the self-managed server room

The meeting quite liked having "done" actions still easily accessible for reference, so the list wouldn't be pruned yet.

Report from Computing Executive Group

Reports from units.

• Infrastructure.

  4. AT server room

     We're expecting the following fibrechannel kit: some discs, some AFS servers and some VM servers, but probably not enough to fill a whole rack. There are also some machines in AT which could perhaps be moved to the Forum (where we now have spare rack space again) if necessary.

     Conclusion: add FC kit from the bottom up, and add non-FC kit from the top down, and keep an eye on it.

     In general, machines which are purely for teaching or purely for commercialisation should go in AT.

• Managed Platform.

  2. SL5.4 updates reboot

     There are rather a lot of servers still to be rebooted. All units to push on with this. The VM servers will still need held back to 5.3, but hosted machines can go to 5.4.

  3. Logging

     The fact that the mail log name doesn't currently match the component ("mailng") causes problems with the permissions on the log file. The easiest fix is probably to change the name of the log file.

  • It was noted that the directory limit for AFS filenames is 64K slots, not names. Files with long names can take more than one slot. (One of the repository mirrors had run out with only about 30K files in it.)

• Research and Teaching.

  • It's still hoped that the removal of /platspec can be completed by the end of March.

• Services.

  4. No news is bad news

     Tim will take the impending demise of the news service either to Stuart or to the Teaching Committee (by email).

• User Support.

Topics for discussion

• End of News

  This was covered as part of the Services Unit's report.

• Procedures for shutting down self-managed servers

  This was raised in order to clarify the action on "all". Some ideas:

  • We could allow users to register phone numbers, to which we would send text messages in case of problems. (The intention is already to set this up for COs and managed machines.) There's a "Monitoring system enhancements" project to which this should be added.
  • We should suggest to users that they should configure their machines to shut down cleanly when the power button is pressed. The mp-unit will document what's required.
  • Procedures need to be documented and published, with a front page somewhere to link to procedures and documentation. The inf-unit will arrange this.

  That original action will be replaced by these new ones.

• The future of DICE

  Roger raised the question as a result of anecdotal evidence.

  • There might be a perception that "DICE" is just the same as our managed Linux desktop, whereas it really includes all the infrastructure and services too.
  • The desktop is actually relatively cheap to layer on top. It's mostly RaT-provided and tends to be user-driven.
  • The refresh (or not) of the desktop can be an issue. For servers it tends to be less of an issue, provided patches remain available.
  • Teaching still mostly uses DICE and the desktop. There's still a need for a teaching platform, and this also has to be easily available to teaching staff.
  • Research has migrated to some extent to self-managed machines and laptops for various reasons.
  • Could we do a "managed laptop" of some kind? What would it be? Could we use EE's technology?
  • We don't want to get into the situation of having "semi-managed" machines. That's a support nightmare.
  • AFS probably isn't suited to all purposes, and we will still need NFS in some cases. How do we manage this if we wind down amd support? How do self-managed machines fit in?

AOCB

The next meeting will be on Wednesday 24th March in room 4.31.