Changes between Version 8 and Version 9 of GEC24Agenda/EveningDemoSession


Ignore:
Timestamp:
02/24/16 14:19:04 (7 years ago)
Author:
agosain@bbn.com
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • GEC24Agenda/EveningDemoSession

    v8 v9  
    139139''This demo shows !CloudLab, a facility for research on the future of cloud computing. ''
    140140
    141 This demo will showcase several new features of CloudLab, including:
     141This demo will showcase several new features of !CloudLab, including:
    142142
    143143* Status reports (health and available resources) for clusters
     
    161161happen if I scaled up the number of compute nodes?"", and want to add more temporarily to test a theory --- without recreating other experiment infrastructure. Compute or I/O-intensive jobs may be able to opportunistically use additional resources to increase throughput. Moreover, cluster or testbed resource requirements change as user workloads come and go. Depending on cluster design, location, and resource requirements, it may be useful for clusters to share resources, seeking temporary ""loans"" from under-utilized clusters to increase job throughput at times
    162162of high load.  We have developed new ProtoGENI API extensions, server-side (AM) management and policy code, and client tools to manage experiments whose resource allocations grow and shrink dynamically over their lifetimes. These features support not only dynamic experiments within an AM, but also allow the AM's resources to be used temporarily by other, external clusters. To arbitrate and facilitate sharing between both clusters and experiments with different resources, priorities, guarantees, and users, our dynamic experiment management software employs a mix of flexible policy, soft and hard resource guarantees, and a general, cooperative encoding of resource values among cluster management and dynamic experiment clients to promote eager sharing of unused resources. Our demo will showcase both dynamic experiments, and
    163 inter-cluster resource sharing, at several CloudLab clusters. OpenStack cloud experiments at multiple CloudLab clusters will add nodes when they are available, and give up nodes when the local cluster is under pressure. One CloudLab cluster will share its resources with a Condor pool, and the CloudLab share of the Condor pool will grow and shrink. We also hope to have another CloudLab cluster integrated with an HPC cluster running Slurm,with the HPC cluster requesting CloudLab nodes based on its workload demands, or releasing them when CloudLab is under resource pressure (and thus requests or demands them back). We will be able to twiddle policy knobs to induce dynamic change and
     163inter-cluster resource sharing, at several !CloudLab clusters. !OpenStack cloud experiments at multiple !CloudLab clusters will add nodes when they are available, and give up nodes when the local cluster is under pressure. One !CloudLab cluster will share its resources with a Condor pool, and the !CloudLab share of the Condor pool will grow and shrink. We also hope to have another !CloudLab cluster integrated with an HPC cluster running Slurm,with the HPC cluster requesting !CloudLab nodes based on its workload demands, or releasing them when !CloudLab is under resource pressure (and thus requests or demands them back). We will be able to twiddle policy knobs to induce dynamic change and
    164164show how the clusters and experiments adapt. We plan for demo participants to see this resource dynamism and a snapshot of the management software's decisions in a ""dashboard"" web page.
    165165
     
    197197
    198198
    199 ==== Building an End-to-end Slice through Slice Exchange between Virtualized WiFi, VNode, and ProtoGENI ====
     199==== Building an End-to-end Slice through Slice Exchange between Virtualized !WiFi, VNode, and ProtoGENI ====
    200200
    201201