At the Microsoft Connect() event today, the Redmond-based software (and market cap) juggernaut is making a series of technology announcements, covering a broad range of its product and service portfolio.
But, to me, it’s the Azure data, analytics and AI story that gets richer today. That’s because two compelling services on the Azure cloud have each been revamped. And the changes are significant enough to move these services from being merely interesting to being compelling and actionable.
Cosmic rays, and RUs
Let’s start with Cosmos DB. Microsoft’s geographically distributed NoSQL service has been technically compelling for quite some time, but the price of admission has been a blocker for many.
Also read: Microsoft debuts Azure Cosmos DB, a superset of its DocumentDB service
Also read: Inside Microsoft’s Cosmos DB
A significant number of Cosmos DB’s new customer targets are coming not from the comparably-priced Google Spanner and Amazon DynamoDB cloud database services but, instead, from cheaply-hosted implementations of MongoDB, MySQL and SQL Server. That’s the constituency that saw Cosmos DB’s pricing as a non-starter. Now, instead of telling those folks that they were thinking about it the wrong way, Microsoft adopted their thinking, did some engineering work and is introducing much lower entry-level pricing to match.
Also read: Google’s Cloud Spanner: how does it stack up?
At the container (e.g. table, collection or graph) level, the minimal amount of provisioned throughput is being reduced from 1000 RUs (request units), to 400 RUs — the equivalent of $24/month, down from $60/month. Think that sounds good but not great? The plot thickens.
For customers who split their data into multiple containers and who previously incurred the provisioned throughput minimum multiple times, Microsoft had previously introduced shared throughput — i.e. provisioning RUs at the database-level instead. And the minimum there is now 400 RUs as well, this time down from 10,000 RUs.
This means the entry-level price for a Cosmos DB database with multiple containers is now $24/month, down from $600/month. Moreover, the provisioned throughput (be it container- or database-level) can now be ratcheted up at a more granular level — in increments of 100 RUs ($6/month), down an order of magnitude from the previous minimum increment of 1000 RUs.
And, on top of all this, Cosmos DB lets customers mix and match container- and database-level throughput provisioning, providing additional flexibility. Then there’s the recently-introduced reserved capacity for Cosmos DB — allowing customers to commit to 1- or 3-year terms, in exchange for a discount of up to 65%. Microsoft also offers free levels of Cosmos DB (through the Azure free tier or a special 30-day standalone offer) for experimentation and a free emulator for development work — on local machines, or in Docker containers. The Cosmos DB tent is now much bigger.
The machine is learning
Meanwhile, back at the AI ranch, Microsoft is announcing the general availability (GA) of the Azure Machine Learning service. Mary Jo Foley has covered the GA news. I’d like to talk about what’s changed in this new version of the service.
Also read: Cloud machine learning wars heat up
The original Azure Machine Learning service — now called Azure Machine Learning Studio — was a sort of visual development environment for doing machine learning (ML) work. It was highly technical, yet proprietary and somewhat code-averse, therefore making it unappealing to data scientists, whose knowledge was nonetheless necessary to make the most of it.
The second go-round of Azure ML, involved a desktop application called Azure Machine Learning Workbench, as well as an AI add-in for Visual Studio Code. While this was much more code-focused, the use of proprietary tooling still alienated data scientists, who don’t want to adopt new tools in order to adopt a new cloud.
Also read: Microsoft Build goes gaga for AI: Azure Machine Learning and beyond
But today’s GA offers a further-refined version of Azure ML, which now works from virtually any environment for the Python programming language — the one most popular with data scientists. Whether using the command line, the PyCharm integrated development environment (IDE), Jupyter notebooks, or even Databricks notebooks, data scientists can now use the Azure ML experimentation, data preparation, training, deployment, model management and monitoring facilities, and use a great variety of ML and deep learning algorithm frameworks (including PyTorch, TensorFlow, and scikit-learn) to boot.
Also read: Databricks comes to Microsoft Azure
Those using Visual Studio Code (which, by the way, runs on Mac and Linux — not just Windows) still have access as well. In fact, that access has improved, as VS Code with the Python extension supports loading of Jupyter notebooks and can perform matplotlib visualizations right inside the editor.
There has long been a saying in the industry that Microsoft gets it right with its products in the third release. And with this third version of Azure ML and the third major revision this year of Cosmos DB’s pricing schedule, this appears to be as true in 2018 as it was when Visual Basic version 3 came out 25 years ago.
And, by the way, Microsoft is also releasing a preview of version 3 of the .NET SDK for Cosmos DB today (as well as CORS support and Apache Cassandra API support in the emulator). So there a few more another data points — in the fit-and-finish story of data and AI services on Azure.