- SQL Server 2017 Machine Learning Services with R
- Toma? Ka?trun Julie Koesmarno
- 440字
- 2025-02-20 14:20:34
Microsoft R Open (MRO)
Microsoft R Open is an open source R distribution that is 100% free and generally available. Microsoft has enhanced this R Open distribution with an additional high-performance multi-threaded feature of Math Kernel Library (MKL) that is optimized for vector and matrix-based mathematical and statistical computations; otherwise, this distribution is fully compatible with legacy R objects and R code.
R Open is also compatible with the CRAN repository, GitHub packages, or any other repository, making MRO widely usable. On the other hand, R Open has some limitations. It is memory bound, which means that it can only handle the datasets that will fit into the memory (client) available. Proprietary ScaleR functions (available in the RevoScaleR package) will not be available under R Open and it will run on all SQL Server 2017 editions, except on Express or Express with Tools, whereas Microsoft R Client/Server will run only on the Enterprise or Developer editions of SQL Server 2017.
Microsoft R Client is the same R distribution as Microsoft R Open and it is built on top of 100% open source R version. Microsoft R Client is the first version from the Microsoft family of R versions that introduces the RevoScaleR package (ScaleR functions). A data wrangler, data scientist, or data engineer (or any other profile) who installs this version will have the ability to use the parallelization and multi-threaded computing, as well as the use of proprietary functions from ScaleR.
There are some limitations to the R Client version. The memory will be limited to a local client machine with the same limitations as Microsoft R Open—data must fit into the local memory in order to be computed. The ScaleR functions will be installed alongside this version, but the processing will be limited to only two threads simultaneously (regardless of the technical specifications of the client machine) and to the local disk and CPU speed. Otherwise, any legacy R packages or R code will be compatible.
Microsoft R Client also brings the possibility to change the computational environment, which means that the computational load can be pushed to Microsoft R Server or SQL Server R Services and any HDFS system to achieve maximum performance. Building the ecosystem with many R Clients and one (or a few) R Servers will give a high-performance analytical environment without having the need for Microsoft R Server being installed locally. It is compatible with the following flavors of R Server: Microsoft R server for Linux, Microsoft R Server for Teradata DB, Microsoft R Server for Hadoop, Microsoft R HDInsight, and both versions of Microsoft R Server-Standalone and SQL Server R Services.