As part of our new CI/CD process for rolling out new innovations on SAS Viya, I caught up with Mark Schneider, Advisory Product Manager here at SAS. When we turned to the topic of SAS Grid, my mind was blown. He explained how bright the future of workload management is with SAS Viya and backed it up by sharing this demo video Scott Parish put together:
At that moment, I realized I needed to capture the answers to all my questions and share them with the world. Here’s how our conversation went:
KR: Why is workload management so important to SAS and the computing patterns we employ?
MS: Organizations are moving their analytic and data management workloads to the cloud. In doing so, they can tap into near limitless compute capacity to provide SLAs previously unattainable with traditional data center approaches. However, that compute capacity comes with a cost, and the challenge is to balance reasonable performance SLAs with acceptable cloud infrastructure cost. The efficient allocation of jobs to computing infrastructure falls under the heading of workload management.
KR: How are we handing workload management today and what’s coming for SAS Viya?
MS: In SAS 9, workload management is handled by the SAS Grid Manager offering. It allows for administrators to map different users and types of jobs to work queues, each of which is prioritized to give the appropriate performance level. For example, the response times of interactive sessions might be prioritized over batch jobs submitted in the background. SAS Grid Manager is also used to ensure high availability of mission-critical servers. It’s also used to map specific job types to machines with specific resource profiles including memory and CPU availability.
SAS is bringing forward this same functionality to SAS Viya. Some of the functionality is inherent in SAS Viya’s Kubernetes architecture. Other features will come in the form of a new SAS Workload Management offering which is layered on top of SAS Viya.
KR: You mentioned Kubernetes, why is that not enough?
MS: Kubernetes alone can provide high availability by setting up contracts such that it maintains a specified number of instances/copies of each service. It can also handle scaling in the sense that new Kubernetes pods can be spun up to handle up to a set number of Compute Servers on demand to support the execution of batch SAS jobs. All this has been available since the initial SAS Viya release in November 2020. But what Kubernetes doesn’t handle is advanced workload management capabilities present in SAS Grid Manager including:
- Option sets – specifying job-specific SAS options to be applied to specific users, groups, or roles
- Resource definitions – mapping jobs with system resource requirements to Kubernetes nodes with matching available resources
- Queues – ordering jobs to be processed in a line, guaranteeing throughput in a first-in-first-out basis
- Prioritization – giving certain queues priority over others to provide increased throughput for mission-critical operations
- SAS program checkpoint/restart – honoring SAS programs which have been instrumented to be paused and restarted at specific points in the code, avoiding the need to rerun programs in their entirety
KR: What is the benefit of adding these capabilities?
MS: With SAS Workload Management for SAS Viya in a cloud environment, administrators can more easily scale their SAS capacity, taking advantage of on-demand cloud resources allocated dynamically by Kubernetes. And because SAS Viya is based on Kubernetes, SAS Workload Management is positioned to be cloud-agnostic, allowing for portability from one cloud to the next.
KR: This is tremendously exciting! When will this be generally available?
MS: SAS Workload Management for SAS Viya is tentatively slated to release later this year. This release will focus on parity with SAS Grid Manager. Future releases are planned to start in 2022 to add workload management of CAS jobs as well as to use SAS Optimization to automate the process of resource allocation within user-defined parameters surrounding performance and cost. This optimization will use static and historical job profiling to automatically allocate workloads in highly efficient fashion while honoring these parameters.
KR: Wait, applying our own optimization technology to our own workload management? [mind blown]. What’s the final takeaway you want readers to understand?
MS: [laughs]I know, it’s very cool. In a nutshell, I’d tell them that dynamic allocation of cloud resources, in whatever cloud you choose, using business rules which prioritize mission-critical operations and allow for reduced cloud infrastructure cost through efficient job distribution, are all differentiating capabilities brought to the table by SAS Workload Management, coming soon to SAS Viya.
Well, there you have it, folks. Head on over to the Coming Soon section of our SAS Viya page to learn about other innovations we’re dreaming up.