The announcement this week that Splice Machine is open-sourcing its product has become just the latest reminder that -- in emerging technology markets -- open source is increasingly the rule, not the exception.
Open-source software is one of those overnight successes that's been a decade and more in the making. It's a far cry from the early aughts when Red Hat and JBoss blazed a trail that still has doubters. Arguably, there's still the issue of whether Red Hat, a publicly-traded, open source company, is a Unicorn from a different twist. Nonetheless, today, when we get acquainted to a new startup, one of the first questions that we pop is whether they're open source.
So it was with more than passing interest that we checked out a few weeks back a fascinating discussion led by venture firm Accel Partners. Convened on the premise that open source was becoming the default business model for the software business, the panel was moderated by Accel's big data champion, Ping Li, and it included CEOs of three of the company's portfolio companies: Cloudera, Cockroach Labs (no, they're not a prospective rival to Terminex), and Sysdig.
A good barometer of the progress of open source can be found by comparing notes from Black Duck Software's annual Future of Open Source surveys that date back over a decade. The 2015 survey indicated a doubling of use of open source to run business IT environments since 2010. This year's survey pointed out that the preponderance of open source was in operating systems, database platforms, and development tools, and that in the next two to three years, the hot spots for open source will be cloud, database, and big data. Hold that thought.
While open source companies were once the exception, resolution of licensing from the copyleft mandates of the GPL license to the more flexible copyright of the Apache license cleared the way for vendors to monetize with unique IP.
Not all open-source models are alike. The open-core model that preserves some unique IP is more prevalent, although there remain stalwarts like Hortonworks and Red Hat, which are adhering to the pure 100-percent open-source model. And then there's the question of openness: open-source software is supposed to prevent vendor lock-in. But what if only a single vendor supports the open-source project? The apt metaphor becomes that of one-hand clapping.
At the panel session, Accel made the case for what it termed "Open Adoption Software". It used some dated (2011 to 2012 vintage but likely still valid) comparisons of downloads between Splunk, delivered to market via the traditional proprietary model, and the open-source Elasticsearch showing a 200x+ difference. Admittedly, the comparison was a bit apples and oranges in that Splunk is a search engine specifically for log data, and Elasticsearch is a more generalized search product that includes a log datastore. While free downloads are hardly exclusive to open source, the opportunity for getting the full product is. Point taken: open source tends to open the floodgates.
Accel also contended that open source is more in tune with customer desires for agility and the freedom to actually touch the code. We'll buy into part of that. Customers like the visibility that publicly-disclosed, open-source project roadmaps provide. And if the open-source project is a community endeavor, theoretically, it should represent the will of the consensus, politics notwithstanding. But as to touching the code, that's only a real benefit for the relative few organizations with deep enough IT departments that have developers willing to monkey around with the code off hours.
The context of Accel's case for open source heavily emphasized the community focus. But not all open-source projects are community affairs as we observed well over a decade ago. In Accel's own portfolio, you have the polar opposites of Cloudera, which has a product that relies largely on Apache community projects, and MongoDB, which leads its own open-source project hosted on GitHub.
In contrast to the community model, the company MongoDB largely controls the agenda of the MongoDB open-source project. OK, reality is not so black and white. As MongoDB has embraced a modular architecture for storage engines, this vendor-led model has loosened somewhat. And even among community-focused open source players, like Cloudera, projects will typically gestate under the vendor's control and might ultimately move to Apache.
When it comes building the business, open source and proprietary are the same -- but different. The biggest difference is starting points. The proprietary software company starts with an idea that is refined based on identifying customer pain points and classic gap analysis. With open source, the trigger is less formal, because at the outset, the primary risk is sweat equity. Somebody gets an idea, develops it in the wild, and in place of gap analysis, there's the sink-or-swim process of developer interest going viral. But, ultimately, both need to deliver some unique value-add, scale it, and go to market.
There is the neatness, or lack thereof, of the open-source model. Witness the long tail of adoption of Android updates, or the ordered disorder of the Hadoop platform, where each commercial platform has different mixes and matches of open-source projects. But there is a parallel with the proprietary world: vendors differentiate, and just because they go open source, why should that strategy differ?
The precursor was standards. SQL became the de facto and formal standard for enterprise databases, yet each major player varied in how they implemented SQL and which set of ANSI SQL features they supported -- not to mention all the ancillary stuff (like storage, tuning, and stored procedures) that make databases, databases.
Open source has a different pricing model, sort of. You pay by subscription rather than perpetual license plus support. But the subscription model is hardly exclusive to open source; it's the predominant ticket for anything in the cloud as well. Nonetheless, open source has impacted pricing expectations. Nobody's going to pay a lot for this muffler -- but that has more to do with the nature of software that lends itself to the open-source model: the software that serves a sufficiently commoditized function that has a large enough addressable market. Open source just doesn't work as well when the software has narrow appeal. The community is too small, and for open source developers, the potential rewards too meager.
What's interesting is that, with a few exceptions (like SugarCRM), open source has not penetrated the application level. The migration activity that's occurring in applications is not necessarily to open source, but to the cloud. It's more critical for enterprise developers to tweak the configuration of the software than monkey around on the source code.
And even within infrastructure, the open-sourcing of the database stack remains a work in progress. Household name databases like Oracle, IBM, Microsoft, and Teradata won't go away anytime soon, and neither will their proprietary software models. There is too much IP -- ranging from how the database is physically to run to how to embed business logic -- to make open source make technical, not to mention economic, sense.
The open source question is over the future of emerging platforms, and here the answer is hardly a slam dunk. In the cloud, there is OpenStack in its many open source implementations, but the presence of an open source alternative is not likely to prod market leader Amazon (its core value is the ubiquity and depth of its cloud stack) to go that direction anytime soon. From a technology stack perspective, Amazon is the cloud equivalent of Oracle; if you use Amazon's Redshift data warehouse with Amazon's database migration tools, S3 storage, Kinesis streaming, Lambda code provisioning, and EC2 compute, it's going to require classic platform porting to get that mix of capability from Azure or SoftLayer.
For data platforms, the open source poster child is Hadoop. But the caveat is that Hadoop is not just open source, but a competing mix of overlapping open-source projects from vendor to vendor. While Hadoop stacks are generally similar, the mix of supported open source and proprietary content varies. For instance, Hortonworks Ranger and Cloudera Navigator are both open source and target similar security and governance tasks, but their coverage extends to varying mixes of components. You can always go from one Hadoop platform to another, but if you use some vendor-specific components, you're still looking at porting.
As for best of breed among the NoSQL operational databases and the MPP NewSQL analytic platforms, the long-term trend is skewing toward open source with the qualification that, if the product is extremely niche, the significance (or benefit) of the strategy is more that of business model rather than one of community.
So, back to our core question: has open source become the default business model for enterprise software? We hate to say this, but the nuanced answer is: it depends.