Science2 publishers2 min readPublished Updated
Microsoft and Google back Apache Ossie's shared format for moving semantic models between platforms
Microsoft and Google are joining Apache Ossie, an open format for moving semantic models between platforms that more than 60 companies already back. Its shared hub cuts converter work, though analysts say lock-in eases only as far as the format can carry each vendor's logic.
The Scientist · Science desk
What happened
- The project began as Open Semantic Interchange and became Apache Ossie when the Apache Incubator accepted it in June.
- Ossie describes semantic models, including datasets, fields, relationships, metrics and AI context, in JSON and YAML.
- Microsoft is building a two-way converter between Power BI semantic models and Ossie, the company wrote in a blog post.
- Microsoft also wants the spec to support more of its ontologies and to recognize DAX, the Power BI language for calculations and business metrics.
- Google is still in the process of joining, a representative said by email, and has not detailed what it plans to contribute.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- cost Teams that move analytics workloads between platforms would spend less effort rebuilding the same semantic definitions each time, according to Aditya Ranjan, a senior data engineer at H-E-B.
- decision Buyers weighing a switch of BI or data platform gain leverage if the move no longer means rebuilding the semantic layer from scratch, HyperFrame Research's Stephanie Walter said.
- capability Developers building AI agents could give every agent the same metric definition across platforms, so two agents are less likely to read one metric differently, Walter said.
Ossie is built as a hub. Each vendor needs one converter to and from the common format, so no pair of platforms needs its own translator [6]. InfoWorld names five platforms the format is meant to work across: Snowflake, Databricks, Tableau, ThoughtSpot and Sigma [5]. Linking those five pair by pair would take 10 two-way converters. Through the hub it takes five [1]. Add Power BI and the counts become 15 and six [2].
The thing the converter count doesn't tell you is how much of a model arrives intact. In a hub design, a model moving between two vendors goes through two conversions, out to Ossie and back in on the far side [3]. Stephanie Walter, practice leader of AI stack at HyperFrame Research, said portability does not necessarily mean complete interoperability [16]. How far a model can move, she said, depends on how much vendor-specific logic Ossie can represent and how accurately the destination platform can interpret it [16].
Microsoft's push to have DAX recognized goes at the first of those limits. The company wrote that enterprises could then carry the calculation logic that gives a semantic model its business meaning along with the model itself [9]. Walter's second limit still applies. A destination platform would have to interpret those calculations accurately [16]. Google's dialect is already there: the specification lists BigQuery/GoogleSQL as supported, a result of "early community contributions that recognize BigQuery's footprint across enterprise data stacks," a Google representative said [11].
The benefits analysts describe rest on the same coverage question. Ashish Chaturvedi, executive research leader at HFS Research, said recreating and maintaining the same business definitions across platforms often leads to metric drift [13]. With Ossie, he said, developers can define a metric once and treat it as a versioned, reviewable code artifact [13].
I think the analysts' hedge is correct. They said the backing makes interoperable analytics platforms a little more likely and is no guarantee that vendor lock-in will go away [2]. The converter savings follow from the design [1]. Whether lock-in eases depends on the vendor-specific calculations. Those move only as far as the format and the destination platform allow [16]. Michael Leone, principal analyst at Moor Strategy and Insights, said the bigger advantage of the backing is that making semantic models more portable gives enterprises more ownership of them [17].
What to watch
- Whether the Ossie specification adds DAX as a recognized query language, and whether non-Microsoft platforms can interpret DAX-defined metrics accurately.
- What Google says it will contribute once it finishes joining the project.
- How much of a Power BI model survives a round trip through Microsoft's two-way converter once it ships.