Microsoft R Server has just released a feature-packed version 9.0.1. Customers can download from MSDN from this link. Enterprise customers with EA agreement will be able to download from Microsoft Volume Licensing Software Center in January 2017.
There are major enhancements and brand new face lifts in some of the capabilities in Microsoft R Server that worth mentioning: -
- Complete redesign of deployment capability in R - Operationalization
Based on the feedback the R team heard from customer, the team did a complete redesign of DeployR with simplified deployment service and using Swagger-based API for application integration. The team renamed the capability to 'Operationalization' which is closer resemblance to the functionality that it represents which is to help customer to deploy and consume the R model that is created.
Operationalization supports many of enterprise readiness features, for example supporting active-active recovery in web nodes out of the box and SSL/TLS 1.2 encryption and AD LDAP or Azure AD authentication.
There are several major differences between Operationalization in Microsoft R Server 9.0.1 and DeployR in Microsoft R Server 8.0.x and earlier versions.Operationalization has the following differences:-
a. Built on .Net Core rather than Apache Tomcat
b. No separate installer for Operationalization - install through Microsoft R Server setup. Configuration is done using Administration Utility.
c. Does not support MongoDB. By default, it installs SQLLite as the metadata DB, but it supports PostGreSQL, SQL Server, or Azure SQL Database if you the admin choose to replace database with these platforms.
d. No longer needs to upload model to repository DB, you can publish the model directly using mrsdeploy package or from a REST API.
e. No more using RBroker framework for API integration, it is replaced with Swagger based API.For complete set of differences, please see Comparing Operationalization in R Server 9.0 to DeployR 8.x.
Some gotcha(s) that you need to take note as well:-
a. Because of .NET core dependency, Operationalization can be installed on Windows Server 2012 R2 and Windows Server 2016, CentOS/RHEL 7.x, Ubuntu 14.04 and 16.04. SLES is not supported until 2017. You can install Microsoft R Server 9.0.1 on CentOS 6.x, but Operationalization capability will not be able to initialize.
b. Configuration is done in Admin Utility instead of Admin Console in Microsoft R Server 8.x.
c. If you currently have R model deployment setup using DeployR 8.x, you need to re-write the deployment using Operationalization in Microsoft R Server 9.0.1 as the underlying API and metadata structure are completely new. There is no migration path available from DeployR to Operationalization in Microsoft R Server 9.0.1.
- "Fast" machine learning algorithms from MicrosoftML package
Some of the exciting algorithms that are high performant and multi-threaded like Fast Tree, Fast Forest, Fast Linear Model and Neural Network are introduced in MicrosoftML package. Some of these execute off disk, and can scale up to multiple hundred GBs in a single node.
MicrosoftML package is available in Microsoft R Server 9.0.1 on standalone Windows, R Client for Windows and SQL Server VNext, it's currently not available in Microsoft R Server 9.0.1 on Linux, Hadoop, and Azure HDInsight.
To view MicrosoftML functions, use R Console with Rgui.exe or R Tools for Visual Studio - help(package="MicrosoftML")
- Support for Spark 1.6 and 2.0
Some important references:-