| Abstract |
Vectors Symbolic Architectures (VSAs) are distributed representations that combine random patterns, representing atomic symbols across a hyper-dimensional vector space, into new symbolic vector representations that semantically represent the component vectors and their relationships. In this paper, we extend the VSA approach and apply it to decentralized workflows, capable of executing distributed compute nodes and their interdependencies. To achieve this goal, services must be discovered and orchestrated in a decentralized way with the minimum communication overhead whilst providing detailed information about the workflow - tasks, dependencies, location, metadata, and so on. To this end, we extended VSAs using a hierarchical vector chunking scheme that enables semantic matching at each level and provides scaling up to tens of thousands of services. We then show how VSAs can be used to encode complex workflows by building primitives that represent sequences (pipelines) and then extend this to support full Directed Acyclic Graphs (DAGs) and apply this to five well-known Pegasus scientific workflows to demonstrate the approach. |