SAS Viya is Now Available on IBM Power Systems
IBMer Beth Hoffman explains why having the right infrastructure to support SAS Viya is important.
By Beth Hoffman03/02/2020
In late 2019, SAS Institute delivered SAS Viya 3.5, which includes its debut for support on IBM Power Systems*. The Power Systems platform is designed for analytics, machine learning/deep learning and AI—workloads typical for SAS Viya—and brings compelling advantages to enterprise clients struggling with commodity hardware. SAS Viya 3.5 is supported on Red Hat Enterprise Linux* on POWER9* servers, including both PowerVM* virtualization-based systems and accelerated servers.
SAS Viya is a cloud-enabled software framework that contains an open analytics engine to deliver quick, accurate and reliable business insights for complex analytical challenges. It’s designed to process huge amounts of data and perform complex analytics. SAS Viya supports developing applications in modern programming languages such as Python, R, Java* and Lua. Sitting on top of the SAS Viya engine is a set of SAS analytics products. Some of the more popular products are SAS Visual Analytics, SAS Visual Statistics, and SAS Visual Data Mining and Machine Learning (VDMML).
Addressing the Analytics Lifecycle
Across the analytics lifecycle—from data discovery to model development and deployment to operationalizing analytics—SAS Viya addresses a wide range of compute demands from the diverse set of applications that run on it. Clients that choose to run SAS Viya 3.5 on IBM Power Systems servers may benefit from superior performance, reliability and infrastructure flexibility.
The IBM POWER9* processor is a leader for performance-critical applications, as shown in its server benchmark results. It enables the throughput that’s critical to SAS analytics workloads with 2x better I/O bandwidth1 and 1.8x larger memory bandwidth2, which are key for SAS Viya’s parallel loading of data and analytic processing. The POWER9 processor uses simultaneous multithreading to boost performance without the security concerns typical of x86.
The ability to scale and deploy more threads per core allows the Power Systems platform to tackle a mix of computing demands at the scale and performance required—helping clients meet their service-level agreements. For SAS Viya products such as SAS VDMML that can leverage GPUs for processing, the CPU-GPU interconnect speeds on POWER9 accelerated servers can be up to 5.6x faster3 with the built-in NVLink 2.0.
More specifically, IBM Power System AC922—the GPU-enabled system designed for enterprise AI—is engineered to be a powerful AI training platform accelerating insights from SAS analytics. Together, this collection of POWER9 capabilities enables superior performance for SAS Viya workloads along with the enterprise-grade dependability the Power Systems platform is known for.
POWER9 servers provide a spectrum of flexible deployment options to align with a client’s SAS Viya workload requirements, infrastructure skills and existing POWER environments. SAS Viya on Power deployment options include running workloads on bare-metal or virtualized servers, on CPUs or GPUs, on-premise or in POWER9 processor-supported clouds, and with internal or external data storage. Figure 1 (below) shows a summary of the key deployment options using PowerVM virtualization-based and accelerated servers and the use cases each are ideal for.
Deployment Options for POWER9
SAS Viya itself has some deployment options that align nicely with POWER9 environments; it can run as a stand-alone tool or can coexist alongside traditional SAS 9 environments (using SAS 9.4 and SAS Grid Manager). SAS 9 code and models can be executed in SAS Viya or SAS Viya results can be brought into existing SAS 9 environments. In either case, clients running SAS 9 on AIX* can choose to deploy and manage SAS 9.4, SAS Grid Manager and SAS Viya on the same POWER9 enterprise-class system within a set of PowerVM partitions (Figure 1, above).
Another option is to co-locate SAS Viya on the same POWER9 server with a primary data source. Co-location provides many benefits:
- Simplified management of multiple SAS environments from one management interface
- Easily scale by leveraging unused resources and adding more SAS Viya partitions as growth occurs
- Simplified networking with SAS Viya nodes and data sources coexisting on the same physical server
- Flexibility to add one or more GPU-enabled servers later when a first use case for SAS VDMML arises
The Right Infrastructure
Having the right infrastructure to support SAS Viya is more than just the servers. A robust storage subsystem will have a measurable impact on the total solution performance. Power Systems servers coupled with IBM Storage systems empower SAS environments with a complete infrastructure that doesn’t make tradeoffs between performance, cost and reliability. The agile, full stack solution accelerates insights, controls costs and maximizes resiliency of SAS Viya environments with ultra-high bandwidth and the highest level of availability. To learn more about IBM Power Systems for SAS Viya, read ibm.co/2vhcqgk.
1 PCIe Gen4 provides 31.5 GB/s throughput versus PCIe Gen3 throughput of 15.75 GB/s based on publicly available material
2 1.8x bandwidth is based on 230 GB/sec per socket for POWER9 and 128GB/sec per socket for x86 Scalable Platform Intel product brief: intel.ly/2FiNAON
3 5.6x I/O bandwidth claim based on CUDA H2D Bandwidth Test conducted on a Xeon E5-2640 V4 +P100 vs Power9 + V100 (12 GB/s vs 68 GB/s rated)
For details on SAS Viya, visit: sas.com/viya
Beth Hoffman is an executive IT specialist and solution architect in the IBM Cognitive Systems ISV ecosystem technical development organization.
Post a Comment
Note: Comments are moderated and will not appear until approvedcomments powered by Disqus