AAASwitch_proposal_Virtualization

Minutes on meeting held at FGCz 18.02.2010

Participants: Can, Christian, Sergio, Riccardo, Alessandro, Peter, Guilherme

We agreed to go for a smaller project proposal reducing the focus to a prototype-like project

Focus

Allowing users to run jobs in Virtual Machines deployed on local infrastructure

Main technical focus

Provide a solution for a resource provider to run VMs hosting jobs; Cloud would be just another type of resource a resource provider may integrate

Notes

  • 3 - 5 FTEs for the project is too much, limit to 2 - 2.5 FTE

  • accounting system needed especially for using cloud

  • we could use a simplified cost model

  • submit proposal in March, aim to 18 months project

  • need to avoid too much fragmentation of available manpower

    Initial assessment phase (survey existing components in the above architecture)

    Duration: 2 - 3 months

    Candidates: OpenNebula (FGCZ has positive experience with it, but no cross-cloud testing)

    Partners involved: all

    Available manpower: 0.2 FTEs

Architecture

See images taken from the discussion on the whiteboard: [http://fgcz- data.uzh.ch/public/cp/20100218--virtualization/][1]

  1. VM management (quite likely already done -- survey existing solutions)

    administer infrastructure

    store VMs

    deploy VMs

    monitor VMs

    starting from the outcomes of the survey, analyze what has to be integrated/implemented

    This should also go hand in hand with the supported deployed and usage scenario

    Involved partners FCGz, UZH/GC3, Switch, UZH/CSG

    Available manpower 0.8 FTEs

  2. LRMS integration

    start/stop VMs on demand

    possibly use cloud resources

    Involved partners: FCGz, UZH/GC3

    Available manpower: 0.6 FTEs

  3. Accounting

    derived from LRMS only?

    Involved partners: UZH/CSG

    Available manpower: 0.4 FTEs

  4. Usage policies

    Involved partners: All

    Available manpower: 0.2 FTEs

First estimation of FTE allocation

  • FGCz = 1FTE

  • UZH/GC3 = 0.5FTE

  • UZH/CSG = 0.5FTE

  • Switch = 0.2FTE

Data management problem

Some applications, e.g. uni-prot, need a large data file to process; how do we stage this to the VM?

  • cannot pre-stage to the VM: uni-prot is updated every few days

  • solution has to be robust: otherwise it impacts on manpower

  • cluster file systems depend on the local cluster

  • what to do?

  • need to survey existing solutions (scp, parrot, fuse, NFS, cluster FSs)

  • at the end of the project, come up with recommendations on what solutions works with what application

Note: Ask system-X to provide Wiki space

[1]: http://fgcz-data.uzh.ch/public/cp/20100218--virtualization/ (http ://fgcz-data.uzh.ch/public/cp/20100218--virtualization/)

top