HPE streamlines use of machine learning services with Haven OnDemand Combinations. Vertica release improves performance, adds Hadoop and Spark support.

Hewlett Packard Enterprise announced August 30 at its HPE Big Data Conference in Boston that it’s making its library of machine learning services easier for developers to build into smart, “cognitive” applications through Haven OnDemand Combinations. In a second announcement at the event, HPE unveiled Vertica 8.0, the next release of the company’s high-scale analytical database.

Haven OnDemand is in the white-hot category of machine learning services. (The category is sometimes pegged as cognitive or artificial intelligence technology, but Constellation Research views machine learning as a more accurate description of current capabilities.) It’s a domain that has seen dozens of acquisitions in recent years, led by leading tech companies including Amazon, Google, IBM, Intel, Microsoft and Salesforce.

Hewlett Packard Enterprise Powers Cognitive Apps from Constellation Research on Vimeo.

Derived largely from the Autonomy IDOL portfolio, Haven OnDemand now includes more than 70 APIs in categories including text analysis, image analysis, audio and video analysis, prediction and search. By comparison IBM has roughly 30 APIs and Watson Cognitive Services while Microsoft has more than 20 Cognitive Services APIs.

How will HPE differentiate Haven OnDemand as the big public cloud companies deepen their portfolios? Haven OnDemand Combinations is an early effort to do just that by enabling developers to bring together multiple APIs in composite services that can be saved and reused for fast development.

HPE has introduced a few pre-built Combinations of its own, including call archiving and language-agnostic sentiment analysis. But these Combinations aren’t products so much as starting points meant to be adapted by customers. The essence of Combinations is giving developers the ability to tie together multiple APIs, using a drag-and-drop design interface. From thereon, developers can quickly invoke the Combinations through a single API with minimal coding.

Haven OnDemand runs on Microsoft Azure, but it’s REST-based APIs can be invoked in any services-enabled environment, including Amazon Web Services or hybrid and private clouds. Haven OnDemand services originated on HP Helion, but the services were relaunched on Azure after the Helion public cloud was shuttered in 2015.

The availability of machine learning services on Amazon, Azure, Google and IBM clouds is clearly a threat to Haven OnDemand. But in an onstage interview at this week’s conference, Microsoft executive Mike Schutz said the first priority for Azure is providing quickly deployable and scalable infrastructure services and compute capacity. Schutz described Haven OnDemand as a “higher-level solution.”

HPE executives say Haven OnDemand has a head start on delivering machine learning services and stressed that they’re already in production use among “hundreds of thousands of users.” The Haven OnDemand community has some 18,000 developers and the APIs are getting “millions of calls” per month, they said. HPE also noted that Haven OnDemand covers essentials for developing secure applications such as role-based permissions for access to data.

Haven OnDemand Combinations gives developers the ability to chain together APIs and
associated data pipelines and invoke them through a single API call.

MyPOV: There’s no doubt that competition from the big public clouds will present perception and performance challenges to Haven OnDemand over the long haul. For starters there’s the one-stop-shop appeal of using development services and machine learning services from one and the same cloud. Even in the case of Azure, Haven OnDemand is a stand-alone site (HavenOnDemand.com) rather than a library of APIs that’s exposed within the Azure Cloud.

On performance, it will be challenged when high-scale used in an app is in one cloud, say Amazon Web Services, while Haven OnDemand Services run on Azure. HPE execs said the Haven OnDemand APIs typically work with indexes and result sets of data that are a shadow of the size of the original data, minimizing performance and storage penalties, but data movement at high scale can’t help but tax performance.

In short, HPE’s competitive advantage will hinge on just how aggressively and successfully HPE and its biggest rivals pursue machine learning services and support for building smart applications.

Vertica 8.0 Bolsters Hadoop, In-Database Analytics and Cloud Support

HPE’s Vertica analytic database is a popular choice for high-scale data mart deployments, data warehouse optimization scenarios and in embedded, OEM use by software and services companies, such as current customers Domo, GoodData and Looker.

The Vertica 8.0 release announced at HPE’s Big Data Conference is due out by the end of October, and it promises better performance as well as extended support for Hadoop, in-database analytics and cloud deployment. Vertica 8.0 is said to deliver faster data loading, simplified data loading from Amazon S3, and comprehensive visual monitoring of Apache Kafka data streams.

HPE is bolstering Vertica compatibility with Hadoop with support for the Parquet file format. Parquet is widely used on Cloudera deployments. Last year Vertica gained support for ORC, a file format often used in Hortonworks, so HPE is rounding out its ability to work with leading Hadoop distributions.

On the in-database front, Vertica 8.0 gains R-based machine learning algorithms that will enable data scientists to model against vast data sets relying on the power of Vertica’s massively parallel processing (and thus avoiding moving data to analytic servers or relying on sampling techniques). The upgrade also adds a two-way connector for Apache Spark so data scientists can rely on machine learning algorithms in Spark or port high-scale queries that might choke the memory space from Spark to Vertica.

Vertica was already certified to run on Amazon Web Services, but the 8.0 release adds support for deployment on Microsoft Azure. On either cloud it’s a bring-your-own-license (BYOL) approach, but you can spin up the Vertica Community Edition from the Microsoft Azure Marketplace.

MyPOV on Vertica

As with Hadoop, as-a-service offerings seems to be the hottest database deployment choice of late, but HPE insists that its customers prefer to manage their own deployments BYOL style. The company took a stab at database as a service with Vertica OnDemand on Helion, but unlike competitors like Teradata and Oracle, it has since eschewed providing managed services.

Certification on Azure is a good step and I won’t be surprised to see deeper ties with Microsoft and perhaps more cloud deployment options alongside Haven OnDemand services. HPE and Microsoft have many joint customers and partners, and it’s exactly those constituents it appealed to in a keynote slide with the simple headline, “Our platform, your vision.” It’s about putting Haven OnDemand and Vertica inside customer and partner apps, and that story only gets stronger when there are plenty of flexible deployment options.