SAS SpeedyStore represents a paradigm shift in how organizations can deploy and operate SAS Viya. By replacing traditional file-based data architectures with a unified, high-performance database platform, SpeedyStore can deliver substantial infrastructure consolidation.
This fall, I assisted a customer in transitioning from SAS 9.4 and SAS Viya 3.5 to a unified SAS Viya environment with SpeedyStore. The SAS 9.4 setup supports use cases involving SPDS, SAS Data Integration, SAS Enterprise Guide, and STP web applications, while SAS Viya 3.5 is utilized across the enterprise for business intelligence with SAS Visual Analytics.
When comparing the infrastructure need with and without SpeedyStore, the SpeedyStore alternative is reducing vCPU requirements by up to 67% and RAM by up to 65%. This dramatic reduction translates directly into lower infrastructure cost, and third-party licensing costs for platforms like Red Hat OpenShift and VMware vSphere, potentially saving hundreds of thousands annually and simultaneously improving performance, simplifying data management, and enabling real-time analytics capabilities.
Building upon the concepts introduced in my previous posts on Enhanced Compute Environment (ECE), this analysis explores how SpeedyStore complements ECE's CAS-optional strategy to deliver a more cost-effective, scalable, and operationally efficient SAS Viya deployment.
Traditional SAS Viya Architecture
Traditional SAS Viya deployments rely on a file-based data architecture where Cloud Analytic Services (CAS) serves as the primary in-memory analytics engine. Whilst CAS delivers exceptional performance for large-scale, multi-user analytics.
Multiple CAS node pools handle different workload types (controller, workers), each requiring dedicated storage, memory, and compute resources. Data must be loaded into CAS memory from various sources (sasbdat, sashdat, spds, flat files, RDBMS, cloud repositories), leading to duplication and increased storage costs.
As discussed in my Enhanced Compute vs CAS article, each SAS session requires its own compute instance with dedicated SAS WORK storage. When multiple users process the same data, users are duplicating the same data. This is not an issue, and it works as intended, but it can be architecturally problematic when the datasets in SAS WORK are large.
Universal Storage for AI and Analytics
What Is SpeedyStore?
SAS SpeedyStore is SAS's next-generation unified platform for transactional and analytical workloads, built on SingleStore's distributed SQL database engine and deeply integrated with SAS Viya. Announced at SAS Innovate 2025, SpeedyStore (formerly "SAS Viya with SingleStore Premium") represents a fundamental shift in how SAS Viya manages and processes data, providing customers with full SQL support for real-time, streaming, and batch workloads whilst maintaining complete integration with SAS Viya's analytics ecosystem.
SpeedyStore supports a unified approach for AI and Generative AI by combining multiple data models in one platform. It offers a native vector data type for storing embeddings and performing similarity searches, alongside JSON for semi-structured data and full ANSI SQL for structured workloads. This enables hybrid queries that mix vector search, semantic filtering, and transactional analytics. Integration with LLM frameworks and built-in support for RAG patterns, SpeedyStore eliminates the need for separate vector and SQL databases, delivering low-latency, scalable AI applications on a single engine.
Core Architecture Principles
Unlike traditional architectures requiring separate databases for transactional (OLTP) and analytical (OLAP) workloads, SpeedyStore's hybrid storage model combines both in a single system. This addresses the challenge noted in ECE vs CAS discussions where transactional data and analytics require separate databases, eliminating costly ETL pipelines.
SpeedyStore supports both rowstore (in-memory) for fast transactions and columnstore (on-disk) for analytics. The rowstore provides massively concurrent updates, fast lookups for operational database transactions, whilst the columnstore delivers "millions of rows, fast queries" for data warehouse analytics.
One of SpeedyStore's most compelling features for optimizing TCO is its bottomless database capability, which uses object storage (such as Amazon S3 or Azure Blob Storage) for unlimited storage capacity whilst maintaining high performance. This delivers 20-40% storage cost reduction through automatic tiering of hot data to high-performance storage and cold data to low-cost object storage.
Integration with SAS Viya
SpeedyStore deploys a SingleStore cluster within the SAS Viya namespace, consisting of leaf pods and aggregator pods. The SingleStore cluster includes:
- Main aggregator (MA pod: Responsible for data definition language execution, cluster monitoring, and failover.
- Child aggregator (CA) pods: Running data manipulation commands on leaf pods and reading metadata from the MA pod.
- Leaf nodes: Storing data partitions and functioning as compute nodes, with the SAS Embedded Process deployed to each leaf pod.
A big part of the infrastructure reduction is that SAS Visual Analytics can push CAS actions into SpeedyStore. CAS actions like filtering, computed columns, and aggregations automatically offloaded to SpeedyStore for in-database execution. This reduces data movement between CAS and SpeedyStore, improves performance, and minimizes memory usage in CAS by executing operations directly where the data resides. However, not all CAS actions are currently supported for pushdown, so optimization of reports is key to fully realizing these benefits.
Building on ECE
In my previous article "Enhanced Compute vs CAS – Which Fits Your Needs?", I explored how ECE addresses the need for a lighter, more cost-effective option for programming-centric tasks, ETL pipelines, and Python-driven analytics.
The first practical step of making SAS Viya CAS optional was taken in the 2025.12 release when this option was made available for programing only deployments.
How SpeedyStore Complements ECE
SpeedyStore takes the CAS-optional strategy further by addressing one of ECE's remaining challenges, efficient shared data access. As noted in my ECE article, when using ECE, just like in SAS 9.4 you need to extract some data from somewhere to put in your SAS WORK, then do analytics, meaning that there might be many users duplicating the same data.
SpeedyStore solves this through its shared database architecture, where curated data is available to SAS, open-source, and third-party tools simultaneously. This combines ECE's computational efficiency with SpeedyStore's data management capabilities to deliver:
- Compute workloads running efficiently on ECE without CAS infrastructure and in SpeedyStore with
- Shared data access through SpeedyStore's distributed database.
- Real-time analytics enabled by SpeedyStore's sub-second latency at high concurrency.
- SAS Visual Analytics still supported through CAS when needed, now pulling data in real time from SpeedyStore, also pushing aggregations into the database which makes it even more efficient.
CAS is still the backbone for SAS Viya's visual interfaces like Visual Analytics, and you will still need a CAS server to serve the result from SpeedyStore if you have that in your SAS Viya architecture. This hybrid approach of ECE for computing and SpeedyStore for storage, minimal CAS for visual interfaces is representing the optimal architecture for many deployments.
Why the Dramatic Infrastructure Reduction?
Going back to the customer case mentioned in the beginning, several architectural factors contribute to this infrastructure consolidation.
The traditional deployment requires multiple large CAS worker nodes (32 vCPU each) to handle in-memory analytics. With SpeedyStore, this is reduced to just 3 smaller CAS workers (8 vCPU each), as much of the analytical processing is pushed down to SpeedyStore's in-database processing capabilities.
SpeedyStore's SAS Embedded Process, deployed to each leaf node, enables WHERE expressions, computed columns, and even DATA step processing to execute directly in the database. Performing operations in the database reduces the amount of data sent through the network.
The Third-Party Licensing Cost Impact
For on-premises deployments running on container orchestration platforms, these infrastructure requirements directly translate into significant licensing costs, below is based on public pricing of OpenShift.
Based on 2025 pricing models, Red Hat OpenShift Container Platform (OCP) costs approximately $58,000 per 2-socket license, translating to roughly $906 per vCPU per year (assuming 32 vCPUs per socket). The more cost-effective OpenShift Kubernetes Engine (OKE) reduces this to approximately $236 per vCPU per year.
Following Broadcom's acquisition of VMware in 2023, significant pricing restructuring has occurred, with many organizations experiencing substantial cost increases. The shift to subscription-based licensing with 72-core minimums has created additional financial pressure on virtualization infrastructure.
Storage Cost Savings
Beyond third-party licensing and capacity needs like vCPU and RAM, server needs, SpeedyStore delivers substantial storage cost reductions through multiple mechanisms.
For example, a 75%+ compression rate means a 1 PB traditional SAS datasets footprint can be reduced to approximately 250 TB in SpeedyStore. Using Azure to illustrate:
- Traditional: 1 PB on Azure NetApp Files Premium ≈ £3.25m/year
- With SpeedyStore compression: 250 TB ≈ £812,500/year
- With bottomless tiering: 200 TB hot + 50 TB cold ≈ £685,000/year
- Total savings: £2.56m/year (78% reduction)
By automatically tiering cold data to low-cost object storage (Azure Blob, AWS S3), SpeedyStore delivers an additional 20-40% storage cost reduction beyond compression alone.
Complementary Technologies
Combined with Enhanced Compute: As discussed in my ECE articles, pairing SpeedyStore with Enhanced Compute delivers maximum infrastructure efficiency:
- ECE handles programming-centric workloads without CAS overhead
- SpeedyStore provides shared data access without SASWORK duplication
- Minimal CAS deployment supports only Visual Analytics requirements
SpeedyStore works alongside SAS's open format strategy (Parquet, CSV) for flexible, open data access. Organizations can move more data to Parquet for additional use cases.
SpeedyStore vs DuckDB
SAS offers multiple data management solutions in addition to SpeedyStore, each targeting different scenarios. The newly released SAS Access to DuckDB excels for lightweight, file-based analytics on Parquet, SpeedyStore is designed for governed, enterprise-scale environments requiring both OLAP and OLTP capabilities. The DuckDB features has been a topic of many previous posts. The two technologies are complementary rather than compete.
DuckDB’s strengths:
- Serverless, in-process execution (no infrastructure overhead)
- Excellent Parquet performance
- Simple deployment for programming-centric workflows
- Cost-effective
SpeedyStore advantages:
- Real-time analytics and reporting
- High-concurrency operational workloads
- Strong governance, security, and compliance
- Consolidation of multiple data stores
- Ultra-fast ingest and query performance
- ANSI SQL compliance
- Hybrid/multi-cloud scalability
Conclusion: Maximizing TCO
SAS SpeedyStore represents a fundamental shift in how organizations can architect and operate SAS Viya, particularly for on-premises deployments where third-party licensing costs are a significant consideration and a large upfront infrastructure investment could be needed. By complementing SAS Viya with a unified, high-performance database platform, organizations can achieve:
Infrastructure consolidation
The mentioned 67% vCPU reduction and 65% RAM reduction directly translate into lower third-party licensing costs—potentially saving hundreds of thousands annually on OpenShift or VMware licenses. But SpeedyStore can also be used to consolidate other data sources.
Enhanced operational efficiency
Eliminating data duplication, reducing ETL complexity, and enabling real-time analytics capabilities to simplify operations whilst improving performance. The combination of in-database processing, automatic storage tiering, and shared data access addresses fundamental limitations of traditional architectures.
Strategic flexibility
By supporting both transactional and analytical workloads in a unified platform, SpeedyStore positions organizations to consolidate multiple data stores for performance and cost efficiency whilst maintaining the scalability needed for future growth.
Complementary to ECE
When combined with the Enhanced Compute Environment discussed in my previous articles, SpeedyStore delivers the optimal balance between computational efficiency and data management capability.
For organizations planning SAS Viya deployments or modernizing existing platforms, SpeedyStore warrants serious consideration, particularly when:
- Infrastructure and licensing costs are a primary concern
- Real-time analytics capabilities are required
- High concurrency with many simultaneous users is expected
- Data consolidation can deliver operational efficiencies
- On-premises or hybrid cloud deployment is mandated
