Experiment Descriptors in BonFIRE
The BonFIRE concept of experiment descriptor currently consists of three sections although others could be added and it would be made more general. The three sections are:
- Initial resource deployment
- Configuration of monitoring
- Specification of elasticity rules and actions
The initial resource deployment is the specification of the initial set of compute, storage and network resources that are within the experiment. This specification may include some constrains that must be met when planning the order in which the resources must be created.
BonFIRE allows resources to be monitored. We support various types of monitoring: VM, app, infrastructure and network. These could be configured in the experiment descriptor.
The elasticity rules and actions could also be specified in the experiment descriptor.
Release 1 of BonFIRE does not support experiment descriptors. Instead users must interact with the system via multiple OCCI calls that perform CRUD operations on single resources.
Release 2 of BonFIRE will support experiment resources but will only support the specification of the initial resource deployment.
Within my mind at least there is still a debate as to whether users would actually wish to specify their monitoring and elasticity functionality in an experiment descriptor or in the VM images they construct for the experiment. BonFIRE is taking the approach that initially experimenters will have to specify these in the VM images but over time this configuation can move to the experiment descriptor when appropriate.
We are initially looking at OVF as the format of the experiment descriptor. OVF should offer advantages in terms of existing tooling that comes from being an existing standard. Unfortunately it it not user-friendly without appropriate tooling on top so we will also develop a very simple user friendly experiment descriptor document that may be unique to BonFIRE but will allow users to do must of what they wish very simply.
Experiment Descriptors in TEFIS
The TEFIS platform
In TEFIS, our aim is to manage experiments at all stages of the service lifecycle, providing a single access point to manage complex experiments needing resources from multiple, federated testbeds in support of different aspects and stages of the investigation. Experiments are viewed from different aspects. On the one hand, the resources needed to support an individual iteration or test run of the experiment, and on the other, the domain as identified by the experimenter themselves. These two aspects – resource requirements and experimenter viewpoint – have determined how we are attempting to define experiment descriptors.
The objective is to be able to manage the experiment throughout its entire lifecycle and for each individual iteration (test run). In addition, any metadata as well as experimental data are maintained as a resource beyond the lifetime of the experiment itself for the community at large to benefit from. We try to cater for these stakeholders:
- The experiment owner: to manage their contractual, configuration, performance data and test results;
- The experimenter community: to identify similar experiments that can be used as the basis for related
work or as a state-of-the-art reference;
- The TEFIS component (including any external tools in later versions of the platform): to provide the
means to identify and interact with the data associated with an individual test run step (or process);
- The TEFIS testbed: to manage the data needed in support of process execution, including managing the
data between processes especially when run across different facilities; and any long-term output data held temporarily for the experimenter.
The metadata associated with an experiment or individual test run includes identifiers for ownership as well as access control, experiment description, provenance and curation. In the initial version of the TEFIS platform, a small, TEFIS-specific set has been implemented. Already with this version, we can support experimenters looking for similar experiments on the basis of resource as well as free-form description. As we move forward with individual use-cases and experimenters, we are considering existing standards such as Dublin Core and the Full Metadata Format as part of a more detailed descriptor supporting the different experiment stakeholders.
- Initial Telco 10.00 Tuesday 7 June (1 hour max).