This chapter describes the advantages of the use of SAS Viya 3.5 on IBM POWER9.

As an in-memory AI and analytics platform, SAS Viya 3.5 workloads are data-intensive and require high throughputs of communication with memory and external disks. IBM Power Systems is optimized for workloads that process large amounts of data, which provides high bandwidth, performance, and resilience. Those characteristics are-well suited for workloads, such as SAS Viya 3.5, and can provide a more efficient execution environment and accelerate time-to-insights.

This chapter includes the following topics:

Over the years, data is becoming a valuable asset to companies. They use it in analytical processes to get insights that improve business decisions and differentiate them from their competition.

Their data scientists are creating analytical models to help make data-driven and real-time decisions in many areas: understanding customer needs, product features design, market campaigns, and supply chain efficiency.

SAS Institute is a software developer development company that is providing analytical solutions for over 40 years at the time this publication was written. Organizations use those solutions to process massive amounts of data, which generates critical business insights, but places heavy demands on the IT infrastructure.

For many years, IBM developed infrastructure solutions that support those massive data-driven workloads. IBM Power Systems and IBM Storage Solutions are consolidated platforms in the marketplace that were developed to help those types of workloads. One of its key aspects is the broad set of capabilities to offer an optimized infrastructure for analytic solutions, such as SAS Viya 3.5.

IBM Power Systems is an open, secure, and flexible platform designed for data solutions. It has high memory and I/O bandwidths capable of handling massive amounts of data.

After all the traditional SAS analytical platform and now with the SAS Viya 3.5 platform, IBM infrastructure capabilities offer key differentiators for SAS workloads, including traditional and Viya.

IBM and SAS have a long history of collaboration and have been business partners for over 40 years. IBM provides infrastructure capabilities with industry-leading performance, allowing SAS customers to accelerate time to insights.

That partnership history resulted in the following key outcomes:

From the Scale-Out systems, such as the IBM Power System AC922 to the Scale-Up systems (for example, the IBM Power System E980), IBM Power Systems are designed to handle large amounts of data to drive performance for data-driven workloads, such as traditional SAS and SAS Viya 3.5.

Its architecture includes the following key features:

In comparison to x86, the IBM POWER9 processor has four times more threads for high-performance cores. The I/O bandwidth increases more than five times the x86 bandwidth. In addition, approximately 1.8x more memory bandwidth is available per socket. The IBM POWER9 also incorporates PCIe Gen 4 and NVLink 2.0, which supports high-speed connectivity between the CPU and these connected devices by way of these methods.

With these features, the IBM POWER9 CPU is a workhorse for SAS Analytics, which makes it uniquely designed to manage the most complex SAS workloads, which are typically based on massive data movement. You can gain insights faster with superior power performance. Optimizing buffering, task deployment, and look ahead mechanisms result in increased throughput, which results in significant time efficiencies and drastically limits any bottlenecks.

The IBM POWER9 processor quickly adapts with extreme scalability to grow with the workload and absorb more capacity on the installed system (processing, memory, and I/O). The system optimizations and flexibility of resource allocation swiftly adjust on demand. Spectrum Scale drives a significant advantage in I/O bandwidth and overall speed to enhance this unparalleled scalability.

In comparison with x86, the IBM POWER9 chip has significant I/O advantages over x86 chips. Extreme optimizations are available in buffering and task deployment. Look ahead mechanisms also are available that result in increased throughput, which results in significant time efficiencies and drastically limiting bottlenecks.

Therefore, IBM POWER9 can quickly adapt to unparalleled scalability. It does so by growing with the workload and absorbing more capacity on the installed system (processing, memory, and I/O). The system optimizations and flexibility of resource allocation is another way in which this system is scalable. Finally, IBM Spectrum Scale drives a significant advantage in I/O bandwidth and overall speed.

For large, diverse implementations, the overall solution is likely to be cross-platform and cross-technology, which requires a work effort for overlapping teams from a server, network, storage, and application. IBM POWER9 and IBM FlashSystems deliver the robust infrastructure that SAS large block and sequential I/O workload require. These types of hardware infrastructures can be heavily stressed with large-scale SAS workflows and are always able to handle the workload.

The SAS Mixed Analytics workload consists of a mix of jobs that run in a concurrent and back-to-back fashion. These jobs stress the compute, memory, and I/O capabilities of the infrastructure. The SAS test team described the testbed that they used as a good average SAS Shop set of workload mix.

One of the examples is the SAS Mixed Analytics 20-session workload. It has 20 individual SAS jobs, including 10 computing-intensive, two memory-intensive, and eight I/O-intensive jobs. The tests are a mixture of short-running (in minutes) and long-running (in hours) jobs and the tests are repeated to run concurrently and serially to achieve a 20-session workload.

The testing uses a 20-session and a 30-session SAS MA workload. Therefore, for a single node 20-session workload, a total of 71 jobs are run. A similar scaling of the 30-session workload is used in which 101 jobs are running. Some jobs use existing data stores, and some jobs use data that is generated during the test run.

Flexibility, cloud, and reliability are the three main factors that help SAS and IBM POWER9 empower advanced analytics. Regarding flexibility, most SAS Viya projects likely are with clients who use analytics with traditional SAS. They likely also have other mission critical workloads, such as databases (as perhaps SAP HANA), and other data lakes that are running Hadoop or other kinds of big data. It is important to know that most clients are sensitive to unplanned downtime and they do not tolerate their analytics platforms being unavailable for hours. As a result, the analytics are moving to enterprise servers and higher reliability environments.

Therefore, we can colocate on our enterprise server (for example, the E980 or E950) the new workload (such as SAS Viya), which is running on Red Hat Enterprise Linux and traditional SAS, which is running on AIX and even other mission critical data sources (for example, databases).

The ability to colocate SAS Viya and traditional workloads is important and creates flexibility for our clients. We combine this feature with the efficiency of the platform. This ability to be flexible and run PowerVM on our systems gives clients the ability to optimize the allocation of cycles for their dynamic analytics workload.

 

They also can optimize cycles between SAS Viya and other mission critical workloads. The flexibility never is found on a bare metal system or commodity hardware. You can still use bare metal systems, accelerated servers, or scale-out systems; however, for large clients who are deploying mission critical analytics in production, it is likely that they use enterprise servers.

SAS is making a strategic decision to fully focus on Red Hat. SAS announced that SAS Viya is a cloud-oriented product. Starting with the next release of Viya (Viya 4), the product is based on OpenShift and Kubernetes. A client that wants to deploy SAS Viya will move to Red Hat. Because Red Hat is part of IBM, we joined forces to create added value for our clients.

Furthermore, the reliability of the joint solution is a game changer. This feature is a more traditional value proposition, but important for SAS analytics. The ITIC 2020 Global Server Hardware Survey² shows that for the last 10 years, POWER has the best reliability for our clients compared to other competition, which enables the most reliable solution for our clients.

Finally, the joint partnership between SAS and IBM is the underpinning component that helps to hold up these points. Through POWER8 and POWER9, IBM became deeply involved with the development process at SAS R&D. SAS R&D validated that the new generations of POWER were compatible they showed a great influence in adding SAS Viya availability for IBM Power Systems on Linux. The joint testimony from SAS and our clients proves conclusively that we have something that the competition lacks.

A mix of computing demands requires a system that can meet the requirements of the SAS workload. Organization into categories of model building, deployment, and production is another way to accelerate. Putting models into production requires a constant recommendation engine and continuous scoring. Event stream processing requires consistent scoring of data streams (for example, drones capturing imagines and always scoring against the model).

By using IBM Power Systems, you can obtain 1.8x larger memory bandwidth, 2x better I/O bandwidth, 2x better per-core performance, and reduce your organizations’ IT burden with policy-based automation and self-service tools.

The unified solution stack, comprising server, storage, and services, reduces the compute time, controls costs, and maximizes the SAS environment’s resilience with ultra-high bandwidth and highest availability.

Flexibility and choice of deployment that address all business needs from SAS applications helps to provide unity. Now, we can create a unified solution through dynamically scaling compute and memory, on-demand, with cloud deployment. This solution allows for increased automation that is required to operationalize analytics across an enterprise.

As you automate, much more iteration is needed during automated workflows. This need requires scalability to support iterations as new data comes into the pipeline and must be updated. Deploy multiple production workloads and multiple operating systems (for example, IBM i, AIX, Linux) in a server with workload isolation. End-to-end security is built in at all layers, from processor to the operating system, which allows for one integrated enterprise solution.

This robust and scalable system maximizes resiliency because it provides the highest availability non-mainframe Linux platform for over a decade. Therefore, it is providing uninterrupted analytics that are required for SAS Viya 3.5 workloads. It allows zero impact planned maintenance by effortlessly moving workloads across generations of IBM Power Systems.

IBM Power Systems (in fact, the entire IBM stack) is designed and built for big data. Consistently, we can deliver systems with ultra-fast I/O that can handle massive amounts of throughput. This feature was critical in the past with SAS 9 on AIX traditional workloads and even increases in importance with SAS Viya 3.5’s in-memory parallel load approach. The ability to scale and deploy more threads and core allows IBM Power Systems to tackle a mix of computing demands at the scale and performance that is required, which helps our clients to meet their SLAs.

Given the growing focus on AI and ML/DL from SAS, we plan to bring to the market one of our deployment options: the IBM Power System AC922. This option is our GPU enabled system that is designed for Enterprise AI. These systems are the same systems that power the two fastest supercomputers in the world. Couple this feature with IBM POWER9 CPUs with enterprise-grade dependability and you can be sure of our intent to bring the best solution to the market for SAS Viya 3.5.

This section describes pairing SAS with IBM Storage solutions.

For the 30 session SAS MA tests, the solution used an IBM E980 as a baseline server. The server is paired with an IBM FlashSystem, a Fibre Channel (FC) switch that was configured for 16 GB, and an IBM Enterprise Storage Server GL6S that was configured for 100 GB with Mellanox Switches. The results show that these IBM FlashSystems reduce real-time writing and reads/writes much higher, as shown in Figure 2-2.

SAS 9.4 with converged infrastructure file I/O, IBM POWER8 servers, and IBM Elastic Storage® Servers (Enterprise Storage Server) GL4 storage system with Mellanox network switch can configure the architecture as shown in Figure 2-3.