Update: Stardog is our entry into the commercial RDF database market.
There’s plenty of talk about the purely technical aspects of RDF databases but considerably less talk about the RDF database market as a commercial software business. As I see it, the commercial RDF database market contains at least seven systems, listed here in random order:
There are another 8 to 10 technically viable RDF databases available. However, I want to talk about these seven systems as comprising the commercial market, by which I mean systems that satisfy two conditions: first, they are production-ready; and, second, they are commercially licensed (that is, you have to pay money to use them). A weaker version of the latter condition, which I’m happy with, is just that there is a single entity which owns the system (i.e., they are not community-owned) and that entity is commercial (i.e., profit-seeking) in nature.
Most of these systems have a zero-cost version or are open-source licensed. But they also all have commercial editions or commercial add-ons, extensions, etc. (BigData is a variant: it’s using the dual licensing (GPL and commercial) model.) I also know that SDB, TDB, Sesame, and plenty of others are production-ready, but none of them is commercially licensed, to my knowledge—please post a comment if I’ve got that wrong.
A final qualification: the 5Store site says that it “may be possible to license” it from Garlik. My guess (and it’s only a guess) is that it’s more likely if you aren’t licensing it in order to compete with Garlik’s commercial service offerings. I will just assume, however, that 5Store is licensable generally.
I don’t want to quibble about the contours of the definition; rather, I want to speculate about how we might rank these systems in terms of commercial impact; that is, who’s earning the most revenue from license sales of these systems, what are some of the features of the commercial market with respect to messaging and positioning, etc.
No Pure Plays
The first point to make is that there are no pure plays here. BigData comes the closest to being a pure RDF database play; but since it’s currently GPL’d, it may be the case that its development is being subsidized via commercial activities other than software licensing. I hope Systap is licensing BigData and its extensions; but they may well be generating revenue from customizations, installations, or other service revenue.
Every other system is the product of either a larger software enterprise (Virtuoso, AllegroGraph, OWLIM, Talis Platform), a much larger software enterprise (Oracle), or a related business (5Store, which is available from Garlik, a personal information management service in the UK).
I don’t know of any RDF database produced as a pure play biz model. I don’t know what that means, if anything, but it’s worth noting. Put another way: if that’s a viable business model, I don’t know of anyone pursuing it. Again, if you have other information, please let me know.
Most of the systems are integrated into a larger platform or framework of additional offerings. That’s obvious in the case of Oracle; everyone knows that. Virtuoso has a lot of additional tools, libraries, etc., largely focused on Linked Data, but also including a full RDBMS, XML support, etc. Judging informally, I would say Oracle and Virtuoso have the biggest spread of integrated offerings to complement their respective RDF systems. AllegroGraph has a wide variety of analytic systems that are integrated with it, as well as editors, the RacerPro OWL reasoner, etc. OWLIM has integrated NLP and related analytic tools, etc. Talis Platform takes a slightly different approach; it includes an RDF database, but it emphasizes the integrated offerings: the Linked Data publishing platform, etc.
The outliers here are 5Store and BigData. As far as I know, 5Store is an RDF database only, more or less. I am not certain about BigData: it has some extensions (reasoning, temporal, HA); but I don’t know their status.
Focus on Scalability
This one is semi-technical: the marketing of each system explicitly focuses on database scalability as the primary metric of evaluation. The marketing in the RDF database market, such as it is, reminds me a lot of Carl Sagan: billions and billions of stars, er…, I mean triples. The talk is not of transactions per second or average query times for complex OLAP-style queries. The most common metrics are loading time and the raw number of triples. Several of the systems are built around some kind of distributed or clustered architecture, the point of which is scalability (though reliability is also a technical factor, but not one which any system except BigData emphasizes). AllegroGraph is a near outlier here since it’s now being marketed explicitly for OLTP or transactional loads. Talis Platform is also an outlier since it doesn’t especially emphasize scalability in its marketing, which is not to say that it isn’t scalable (I don’t know either way).
Recall: my point here isn’t purely technical; but, rather, how are each of these systems marketed, what is their commercial messaging or positioning, who’s selling software, etc.
Some kind of reasoning is part of all, or nearly all, of the systems in the RDF database market. This is an important point to emphasize, since it sometimes seems that there is a split between the RDF qua graph data structure and RDF/OWL qua knowledge representation camps. That so-called split isn’t really evident when we look at the offerings in the RDF database market:
- Oracle’s system contains a forward-chaining reasoner at (more or less) the OWL2 RL level
- Virtuoso appears to prefer backward chaining over an assortment of OWL and RDFS property types
- AllegroGraph supports what they call “RDFS++ reasoning”, which appears to be pretty similar to Virtuoso’s property-based reasoning; it also supports Prolog rules
- BigData supports RDFS plus some OWL properties
- OWLIM supports RDFS, OWL Horst, and OWL2 RL reasoning
- 5Store is an outlier here: it does not support reasoning of any kind (that I can determine)
- Talis Platform is an outlier, too: it does not support reasoning of any kind (that I can determine)
There aren’t any systems supporting more expressivity than OWL2 RL; but that is what we should expect to see, given the technical affinities between databases, rule engines, RDFS, and OWL2 RL.
License Costs and Pricing Models
Licensing fees and pricing models are a mixed bag:
- OWLIM does per core licensing (900 Euros per core)
- Oracle Semantic Technologies is a free add-on to the Oracle Spatial extension
- BigData: unknown
- AllegroGraph: appears to be licensed per-CPU and list prices don’t seem to be publicly available
- 5Store: unknown
- Talis Platform: Software as a Service model; has both free and for-pay options
SPARQL 1.1 Participation
The RDF database market achieves a high degree of core interoperability: every system implements SPARQL. There may be interoperability issues at the edge of these offerings, since most of those extensions are relatively tool-specific.
That said, it’s interesting to look at which vendors are participating in the ongoing efforts to specify SPARQL 1.1. Talis, Oracle, Virtuoso, and 5Store are active in the SPARQL WG; BigData, OWLIM, and AllegroGraph are not participating publicly in the WG.
What’s the upshot of all this? The market has some continuities (standards-respecting, reasoning offerings pretty uniform, focus on scalability) and some discontinuities (pricing, degree of integration with other systems, business model). It’d be interesting to speculate about which is the commercial leader, in the sense of largest volume of licensing revenue. There do seem to be some parts of the market that aren’t being pursued commercially. I wonder, too, if we’ll see any consolidation or shakeout in this market anytime soon?
What do you think about the commercial RDF database market? Are your use cases and requirements for a commercial system being met by the current offerings?