User Tools

Site Tools


requirements:slice-description

Slice Description

Code SRSM-1
Responsible Ester Lopez
Components testbed server

Description

“Slice description” refers to the need of describing the tesbed slices, so that researchers can run experiments over them. A slice is not only characterized by the resources it is composed of, but also by the connection between them and link characteristics. It is essential for the CONFINE testbed that the slice description mechanism considers the following parameters:

  • Node description: memory, CPU, storage size, Operative System, access to capture and modify the different network layers.
  • Topology: links and link characteristics.

Analysis

Details

There are several factors that influence which are the details (about the slice) needed to be present when describing it:

  • The Slice Description requirement is heavily coupled with the slice request procedure, specifically to the allocation of slivers into actual nodes. There are three different possibilities regarding the allocation of slivers: (1) nodes are chosen by the researcher itself, (2) the CONFINE administrator will decide on which nodes slivers will be placed and (3) an automatic process will compute the best allocation of the slivers into physical nodes. On the first case is absolutely necessary that the information model chosen for the slice description requires for each sliver to associate it with a physical node, whereas on the two second options it is not necessary.
  • On a slice used to run experiments that operates on layer 3, over a virtualized containers in the nodes, slice description will need to define the virtual interfaces needed for each sliver.
  • Some experiments will be highly dependent on the topology of the selected nodes, whereas others will not care, as long as all nodes are accessible from the others. Therefore slice description should support both providing details about the slice topology and ignoring it.

To conclude, the main points to evaluate the different solutions seem to be:

  1. Description of each sliver and its capabilities (computing, storage…)
  2. Slice topology
  3. Sliver allocation
  4. Sliver interfaces

Solutions

There are several information models that could be used as mechanism to represent the CONFINE slices. The ones considered most relevant are discussed below.

Common Information Model (CIM)

The Common Information Model (CIM) is an open standard for representing managed elements on an IT environment and the relationships between them. The CIM schema is split in several areas, among which the ones more relevant for the “Slice Description” requirement are:

  • CORE: describe the core components of the model and their relationships. From this set it is relevant the ManagedElement and the LogicalDevice.
  • Device: defines the devices and its componets, such as Processor, NetworkPort (with its subclasses WiFiPort and EthernetPort), WiFiRadio, Memory and DiskDrive, which are important components on the description of a slice.
  • Network: allows the description of the network componets and topology.
  • System: describes all the components of a system, such as threads, processes, etc. In our case, the only component we are interested is in the class OperativeSystem.

CIM is a very complete schema, which considers almost anything imaginable on the IT subject, however, it goes too much on the detail, making it a hard to use schema. For instance the representation of a DiskDrive is the following:

Network Description Language (NDL)

Network Description Language (NDL) is a network description schema being developed by the University Van Amsterdam. NDL goal is to model the network as functional elements based on ITU-T G.805 and describe them afterwards in a compact syntax. NDL is divided in six schemas (1) Topology, (2) Layer, (3) Capability, (4) Domain, (5) Physical and (6) Technology. Among them, the ones relevant for the Slice Description requirement are:

  • Topology schema which describes the interconnection between the components through relations like “connectedTo”, “hasInterface”, etc.
  • Physical schema: which describes the pysical aspects of the networks such as the interfaces and the links.
  • Technology schemas includes a set of schemas focused on technology-specific schemas, among them IP, ethernet and wireless are the more relevant for our case.

To summarize, NDL is a simpler schema than CIM and the connectivity between components can be described in a easier way, however it is too much focused on network, without taking into account the IT components as storage or computing power, as we require.

Virtual private eXecution infrastructure Description Language (VXDL)

Virtual private eXecution infrastructure Description Language (VXDL) is a language that allows the description of a Virtual Private eXecution Infrastructure (ViPXi), which is a time-limited aggregation of computing and communication resources. VXDL main component is the VirtualInfrastructure (VI), which is similar to the Slice concept in our project. A VI consist on a set of virtual resources, its topology and a timeline which defines the period of time the VI is valid. VXDL covers most of our requirements regarding the slice description and it is simpler than CIM. Some of its componets will have no use in the slice description though (like the route or routingTable concepts). What is missing in this model is on the description of the vNode, a mean of describing its Operative System and which layers can be accessed.

PlanetLab Rspec

PlanetLab RSpec is used to allocate virtual machines to slices. The PlanetLab RSpec describes a request as a list of slivers already allocated on a physical node. As a list of slivers it does not take into account the topology among the nodes selected, but the cpu_share, disk_max and net_max_rate values can be set for each sliver. Details about the operative system cannot be set though.

VINI Rspec

VINI RSpec extends PlanetLab's by adding network information support, therefore links and virtual links (vlinks) can be defined between sites and nodes. A virtual link is defined by its endpoints between involved nodes and optionally by its bandwidth in kbps.

Recommendation

The more appropriate solution seems to be opting either the VINI Rspec or VXDL, since they are the information models more aligned with our requirements for the slice description, and opposed to CIM, their complexity is reasonable and can be tackled. However, since they do not contain all the necessary elements, they should be expanded as needed.

requirements/slice-description.txt · Last modified: 2012/01/26 17:44 by ester