Mapping of QoS parameters: high level – low level
E.g. End user requires an answer in Max time= 5 sec => D-Box decide to execute plugin x inside a cluster, requesting 3 nodes& 6 sec
- A performance estimation process must take place prior the final deployment and execution of the plugins by the end users. :
- Benchmarking of individual plugins. E.g. Plugin idx, executed in a cluster, using 3 nodes, with input abc, takes an average of 3 minutes to generate an average of 100 results
- Benchmarking of composed “pipelines”. E.g. the execution of the pipeline plugin idx + plugin idy + plugin idz,...
- Who must perform this estimation process:
- In case of individual plugins: the plugin developer should provide this info as part of the plugin description
- In case of composed pipelines: the larkc provider?
- Where to store the benchmarking info?
- Central repository?
- As part of the description of every plugin (for individual benchmarking)?
- Learn from historical data (from previous executions)?
- The D-Box must have access to this info in order to decide the plugins to be executed when it receives the query from the end-user