1 / 15

Resource Representations in GENI: A path forward

Resource Representations in GENI: A path forward. Ilia Baldine, Yufeng Xin ibaldin@renci.org , yxin@renci.org Renaissance Computing Institute, UNC-CH. Slicing of a network. Link slivering. Agreements – resource representation cycle. Possibly fuzzy request.

aislin
Télécharger la présentation

Resource Representations in GENI: A path forward

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Resource Representations in GENI:A path forward Ilia Baldine, YufengXin ibaldin@renci.org, yxin@renci.org Renaissance Computing Institute, UNC-CH

  2. Slicing of a network

  3. Link slivering

  4. Agreements – resource representationcycle Possibly fuzzy request Collective possibly filtered manifest Ads from substrate providers Detailed manifest from substrate provider More specific request to substrate provider(s)

  5. GENI Resource representation mechanisms • ‘Traditional’ Network resources • Ethernet links, tunnels, vlans • Edge compute/storage resources • Measurement resources • Including collected measurement data objects • Wireless resources • WiFi, WiMax, motes etc • Lack of agreement on what resources represent will be a significant impediment to interoperability • Agreement on a format is important, but can be dealt with on the engineering level PG Rspec V1 PG Rspec V2 PL RSpec ORCA NDL-OWL OMF

  6. Network resources • Key distinctions • Number of layers • Describing adaptations between layers • Syntax • Tools PL NDL-OWL PG

  7. Aside: why adaptations are critical? • Network adaptations are part of the description of stitching capability • Needed for properly computing paths between aggregates connected by network providers at different layers • E.g. if a host has an interface that has an Ethernet to VLAN adaptation, this interface is capable of stitching to vlans • Consistent way to describe connectivity across layers (tunnels, DWDM, optical) • Also • Important for matching requests to substrate capabilities • E.g. creating a VM is a process of ‘adaptation’ of real hardware to virtualization layer

  8. Network resources: a practical solution • Stay primarily within Ethernet layer • Accept one format to be used between control frameworks • Perform bi-directional format conversion • Only partial may be possible • Hosted services that perform conversion on demand • E.g. NS2/RSpec v1 and v2 request converter to NDL-OWL • http://geni-test.renci.org:11080/ndl_conversion/request.jsp • Works well for several types of links, nodes, interfaces, ip addresses and simple link attributes

  9. Agreeing on wire format • RSpec v2 with extensions is a viable possibility • Conversion fromRSpec v2 is relatively straightforward pending agreement on edge resources • Conversion toRSpec v2 is likely to be partial but probably sufficient for the time being • Nothing below Layer2 is visible to experimenter

  10. Next challenge: Edge compute resources • Ads: • Aggregates can produce (adapt to) different types of nodes • E.g. PL VServer, PG bare hardware node types, ORCA’sXen/KVM virtual machines (and hardware nodes) • Constraints are possible on disk, memory size, number and type of CPU cores • Properties may include location, ownership etc. • Note: internal topology may or may not be part of the site ad. E.g. clouds have no inherent topology that needs to be advertized • Requests: • Based on constraints on type of node, disk, memory size, core type and count, location

  11. Advertising edge resources • A server can be an individual node or a cloud of servers • A site may choose to advertise individual servers/nodes or server clouds • Clouds have no inherent topology, just constraints on the type of topology they can produce and adaptations for nodes • Servers/nodes or server clouds are adaptable to different types of nodes distinguished by • Virtualization (Xen, KVM, VServer, None etc) • Possibly memory, disk sizes, core types and counts, OS

  12. Requesting edge resources • A request for a node specifies multiple constraints on that node • Type of virtualization preferred • Memory, disk size • CPU Type • Core count • OS • Allows policy to pick best sites based on request and resource availability.

  13. Semantic Shortcut examples PlanetLabCluster Produces PL Nodes ProtoGeniCluster Produces PG nodes • Semantic shortcuts • PL node • Virtualiztion: Vserver • Simple PG node • Virtualization: None • CPU type: x86 or ?? or ?? • EC2M1Small • Virtualization: KVM or XEN • CPU count: 1 • Memory size: 128M • EC2M1Large • Virtualization: KVM or XEN • CPU count: 2 • Memory size: 512M

  14. Other considerations • Emerging standards: • OVF – portable appliance images across heterogeneous environments • CF should be able to generate OVF based on combination of request data and substrate information • Higher-level programming environments: • Google App Engine, AppScale • Distributed map/reduce • …

  15. Next steps • Align edge compute resource descriptions • Enable conversion as a GENI-wide service • Test full interoperation PG<->ORCA by GEC11 • Get the conversation started on aligning abstraction models for for • Wireless resources • Max Ott, Hongwei Zhang, ? • Storage (physical and cloud) • Mike Zink, ?

More Related