wiki:DevOps-2012-10-23
Last modified 7 years ago Last modified on 10/23/12 11:51:17

DevOps Weekly Meeting | October, 23 2012

Time & Location: 11am-12:30am in LHL164

Attending

mhanby, tanthony, pavgi, billb, bade, jpr

2012-10-23 Agenda

  • Agenda bash
  • Updates
  • Funding
    • gather data/stats
    • draft slides
    • todo: put research computing day presentations online
  • Large mem nodes
    • still waiting on IB hw
    • nodes having issue with new broadcom cards, need new pxe boot kernel
  • Research storage
    • target: backup of hitachi drives on nas-02
      • Review appasure this weeek
      • set up config with ed to shadow copy luns
        • on hold for space
      • disconnect nas-01 from production luns
    • locate 4 of new systems in hunstville
    • use cases
      • enable user to mount cloud storage or other fabrics to move off data results from scratch (eg. fusermount of S3)
      • enable user to mount disk images on compute nodes for more efficient performance of Lustre with small files (many small files in one big file image)
      • test drdb on virtual drives
  • Lustre status
    • cheaha running at 288 cores
    • recovery-based rebuild is progressing
      • bringing online this week -- production MDT reimage in progress
      • begin working with users to access recovered directory structure
    • todo: analyze rocks boot to see if new driver or different load order on kickstart boot was the problem
  • Questions
    • installing python3 on centos

Summary

A quick review of the agenda. Not much has changed since last week and the items continue to be worked on.

Discussion

Use Python3 on centos, is best to install from source and not an RPM and use python virtual env PythonDevEnv, this will let you have different spaces for python eggs. If it's just you using it you can install it in your home dir. The core python should install fine though you might have issues with dependencies for some eggs, if centos doesn't have the current version.

Swobjects doesn't use autotools, it has a traditional build structure. This just depends on the developer. Also requires a newer version of boost. This is where using the modules framework is helpful because you can cleanly manage environments.

How do you manage modules in general, not just environments? There are tools like lxe and openvz for containers. Solutions like rpath that let you build from the ground up. But this is essentially the role of linux distributions. There are different goals for the curation of specific repositories, eg. CentOS/RHAT are long term stability, Fedora Core seeks to track the latest development. In general, if you are using a cutting edge tool then you want to (or need to) adopt the platform of the developers.