The Benefits of a Mixed IBM POWER9 Environment for SAS Viya 3.5
IBMer Beth Hoffman explains how to leverage POWER9 Infrastructure for SAS Viya.
By Beth Hoffman06/01/2020
Businesses use SAS Viya to generate performance-critical insights on time, every time and help businesses meet critical regulatory deadlines. Operationalizing analytics workflows is an iterative process that involves the processing of massive volumes of data, placing heavy demands on SAS’s IT infrastructure. Therefore, successful SAS Viya implementations require an Agile, resilient and responsive IT infrastructure.
With IBM Power SystemsTM infrastructure, resources can be easily and quickly allocated and optimized where and when needed. Enterprise-class Power® systems that run SAS Viya can scale on-demand, support a set of independent consolidated workloads (e.g., SAS 9.4 in an AIX® LPAR and SAS Viya in Linux® LPARs) and simplify systems management.
Built on POWER9
POWER® servers offer a number of unique deployment options for deploying SAS Viya for meeting these infrastructure goals. One POWER deployment option in particular provides a broad set of benefits with unique advantages: the mixed POWER9™ server environment. This mixed environment uses a PowerVM®-based system as well as accelerated compute servers. The configuration allows clients to take advantage of the scalability and resilience of enterprise Power Systems and GPU acceleration simultaneously. This would be especially ideal for a SAS Visual Data Mining and Machine Learning workload.
SAS Viya includes a complex set of AI and analytics solutions. SAS Cloud Analytic Services (CAS) is the distributed run-time environment component of SAS Viya for data management and analytics. SAS Viya uses CAS in on-premises and cloud environments, in single-machine architectures with symmetric multiprocessing and cluster architectures with massively parallel processing (MPP).
Figure 1 (below) shows a sample SAS Viya MPP deployment in a mixed IBM POWER9 server environment. SAS Viya is installed across multiple types of physical servers, both scale-out accelerated servers and a scale-up, enterprise-class server. In this case, the CAS controller and supporting services would run in multiple LPARs on the E980 system. PowerVM is used as the virtualization layer on the E980. The CAS worker nodes could run in one or more LPARs on the same system (for CPU workloads) and on one or more AC922 servers (for CPU or GPU workloads). The AC922 servers are bare-metal and are not using virtualization.
When SAS Viya is installed in an MPP environment across multiple servers, it’s important to have the servers networked together as a cluster. The following types of networks can be considered in a SAS Viya MPP network design:
- Basic network: This provides users access to the system and provides a path for administrators to perform systems management
- Node network: This is typically a private network that’s used to provide high-speed, high-bandwidth, and low-latency communication between the nodes in the cluster. This needs to be a minimum of 10 Gb network.
- External data network: This provides a separate connection for data transmission between the nodes and the external data source
In the aforementioned example, the E980 is set up using VIOS for virtualized networks and could also be used for virtualizing internal storage. This mixed server environment is using internal NVMe storage devices in each of the servers and LPARs for the workload data, and thus is not showing an external data network. Each LPAR and each bare-metal system has a 10Gbe 2-port network adapter for internode communications. Each node in the cluster also has either a 1 or 10 GbE Ethernet adapter for basic network access.
No matter which type of POWER9 system the SAS Viya 3.5 software is installed on, Red Hat® Enterprise Linux® must be used.
Ideal for Deployments of Any Size
This mixed POWER9 server deployment option is ideal for SAS Viya full deployment environments of any size. The built-in virtualization in IBM Power provides performance isolation and enhanced security. In fact, amongst its major competitors, Virtualization enables businesses to optimize resource utilization and scale server resources on-demand using Capacity on Demand. It also helps businesses reduce their infrastructure footprint and total cost of ownership by consolidating multiple workloads in the same system. This deployment option is also flexible with the type of workloads it can support. Once GPU-enabled workloads are needed, AC922 servers with GPUs can be added to the cluster.
For more information about other deployment options, storage choices and network configuration examples, read the IBM Power Systems for SAS Viya 3.5 deployment guide.
Beth Hoffman is an executive IT specialist and solution architect in the IBM Cognitive Systems ISV ecosystem technical development organization.
See more by Beth Hoffman