A brief look back at SAS Explore 2023
SAS Explore 2023 was held at the Aria Hotel and Resort in Las Vegas in early September. Even though the MGM properties were valiantly working through the aftermath of a hack while we were there, the SAS Explore conference seemed to go off without a hitch. For many of us in the SAS community, this was a first opportunity to see each other in person again. I'm thankful I could go and really enjoyed catching up with so many colleagues and friends.
I delivered a couple of messages for my part while I was there. The first was a presentation, The Intersection of Workload Placement Between Kubernetes and the SAS® Viya® Platform.
The objective was to help bridge the gap for folks with either SAS Viya or Kubernetes experience by focusing on how workload is handled by the various responsible services.
The presentation began with briefly looking at Kubernetes concepts like pods, services, nodes, node groups, autoscalers, labels and taints and then progressed into SAS Viya. In particular we looked at the four default workload classes for SAS Viya and how they're handled.
From there the presentation dove just a little deeper into the SAS Workload Management offering (included for free as part of the SAS Viya platform since August, 2023) as well as how to improve concurrency performance. We also briefly touched on the techniques that the SAS Cloud Analytic Services use to distribute workload across multiple worker hosts.
The next day I delivered a hands-on workshop, Delivering Powerful Analytics Capabilities with the SAS® Viya® Platform and Kubernetes. While the title was different than the presentation, the objective was essentially the same: to help introduce Kubernetes and SAS Viya concepts around workload placement.
This guided exercise provided each attendee with their own personal SAS Viya deployment running in Kubernetes. It began with a tour of the tools and capabilities that a SAS (and/or Kubernetes) administrator would ideally have access to for managing a SAS Viya deployment. For example, we demonstrated how each SAS Studio session requests its own backend SAS Compute Server.
From there, we talked about extending this concept for many users running at the same time and multiplied the concurrency yet again by showing SAS Model Studio running multiple pipelines at once. As covered in the presentation the day before, we modified the environment to better handle the increased concurrency so that the resources in the environment could be efficiently utilized.
The hands-on then switched gears to look at SAS Cloud Analytic Services (CAS). As you know, CAS is the high-performance, in-memory analytics engine that is designed to process the largest volumes of data as fast as possible. And the exercise was to convert CAS from running as a single-node server (SMP mode) to running as a single service distributed to run across multiple host machines (MPP mode).
The trick on that exercise is that the demo environment we were using wasn't intended to run MPP CAS. So we had to hack at the settings and essentially cram MPP CAS to run in an environment that was too small. The lesson there was to show how Kubernetes and SAS Viya configurations work together to keep things running efficiently and effectively. By forcing Kubernetes to allow MPP CAS to run, we could show exactly how those settings worked. The takeaway then is to understand why we configure SAS Viya to run in Kubernetes and how to plan for proper sizing in a real deployment.
I'm looking forward to the next conference coming up in 2024: SAS Innovate. In the meantime, if you're looking for more SAS Explore 2023 content, including the presentation materials, it's now available at:
SAS Communities > SAS Explore >
SAS Explore 2023 Presentations
https://communities.sas.com/t5/SAS-Explore-Presentations/tkb-p/SAS_Exploretkb-board
Find more articles from SAS Global Enablement and Learning here.
