How Can SAS® Workload Management Optimize Cloud Resources ...? Q&A, Slides, and Recording

[#1]

Watch this Ask the Expert session to learn how to get the most out of your cloud infrastructure spending while ensuring the best experience across your entire analytics platform. We’ll show you how to optimize and prioritize SAS workloads within a Kubernetes cluster and how SAS Workload Manager compares to SAS Grid Manager. 

Watch the webinar

You will learn:

• How to optimize the use of cloud resources provisioned to support SAS Viya jobs.
• The Viya replacement technology for SAS Grid Manager on SAS®9.
• How to extend Kubernetes with advanced features like queueing, prioritization, resource management and preemption.

 

The questions from the Q&A segment held at the end of the webinar are listed below and the slides from the webinar are attached.

Q&A

Does SAS App server context also exist in SAS Viya?

In SAS Viya, we don't have the specific term of “SAS Application Server” anymore, but we have the concept of contexts and so a SAS administrator can configure contexts and configure options. Within a context, you can specify SAS options or environment options and you can associate a specific context for specific clients or specific end users. In summary, what we used in SAS Application Servers in SAS 9.4 translates now to contexts in Viya.

This is even more specific because we have a launcher context, but also per-server context: compute context, batch context, SAS/CONNECT context, so you can be more granular in the way you configure the environment. You do not have to, but it gives the administrator more capabilities, more freedom in the way they configure the environment.

See the link to the official documentation at the end of the page.

Is it possible to have different pod templates with different amounts of resources (CPU and RAM) associated to the same queue/launcher?

Usually not to the same queue. It is possible to have different pod templates, yes. What you would do would be to associate each pod template to a specific context and then you associate that context to a specific queue. As of now, it is still a manual process. It's something the Kubernetes administrator must do because pod templates are not managed via the Environment manager interface, they are managed from Kubernetes administration tools. But yes, it is possible to do it.

SAS 9 on Grid requires a lot of space allocated for SASWORK and UTILITY location for all Grid nodes. Does SAS Viya also require SASWORK, GRIDWORK and UTILLOCATION?

Moving to SAS Viya on the cloud requires re-architecting. It requires rethinking your environment and not just moving one to one as it is. Most SAS 9 environments were not distributed, while SAS 9 grid environments were distributed and compute sessions could run on multiple nodes, but now it's the opposite in SAS Viya. Now when you're running SAS Viya, even without SAS Workload Management, it is distributed by default, so the storage considerations that you must apply on SAS Viya are the same, whether you have SAS Workload Management or not. Licensing SAS Workload Management does not change the fact that SAS Viya now is distributed and the fact that we are running on the cloud. Storage considerations on the cloud are quite different than storage considerations on-prem. I would say probably the requirements are lesser, but it requires a more specific design because it really depends on which specific cloud vendor you are deploying SAS Viya, whether you are on Azure, on Amazon, or Google. If you are on OpenShift on-prem everyone can have a different way of managing their storage and so everyone will have different considerations. But again, it is not driven anymore by having SAS Workload Management, but just by being on SAS Viya on the cloud.

What types of jobs can be managed with SAS Workload Management?

It's important to understand we manage only the computer jobs - everything that comes through a launcher from a technical standpoint. This means all the SAS sessions that are started either as a compute server, a batch server, a connect server, they are all managed independently. If you start them from SAS Studio, let's say, or Model Studio, they are all under the umbrella of SAS Workload Management. On the other side, if you send in CAS code that runs in CAS, that's not managed currently.

Can the workload manager distribute between multiple providers e.g. internal and external?

SAS Workload Management will operate within a single SAS Viya node space. So, if the question is, “Can you submit workloads outside of that namespace?” Then the answer would be no because SAS Workload Management runs inside the SAS Viya environment, inside the SAS Viya namespace, and so it monitors everything that is running in the namespace itself. It cannot branch out to something external. It doesn't go out.

If most load is internal, is there an option to overflow peak loads?

Well, that's an interesting use case where we leverage through SAS Workload Management the capabilities that come from the cloud and from Kubernetes itself. I can give a specific example. Let's say you have deployed SAS Viya on Azure Kubernetes Service and you are running some interactive jobs on SAS Studio that use a lot of CPU and all of your nodes are full. If you configure your node pool on Azure to be automatically scaled by Azure, Azure will notice that the CPU load’s gone above a certain threshold. Azure will start a new node and the new node will be in the same computer node pool as the other ones. For this reason, it will inherit the compute label and SAS Workload Management will be able to see, “Oh, now there is one more node available, so now I can start the additional workload there.” So, after Azure scales, SAS Workload Manager can use it without any intervention from an admin or an end-user. It's just there. You can use it. The same works the other way. Once the workload goes down, then obviously SAS Workload Manager will stop sending jobs, Azure will see the load has decreased and will shut down your additional nodes and that's one of the ways, for example, to manage cost and take your resources down and don’t overpay when you don't need it.

What does SAS Workload Management do for me that I could not accomplish just using Kubernetes?

The key point is that it puts all the power in the hands of the SAS administrator so that they do not have to switch between Kubernetes and SAS. The second reason is that Kubernetes is totally unaware of the queues’ priorities that are specific for SAS workloads. Kubernetes does not know if a job is coming from SAS Studio or Model Studio or if it is a batch and so it has different requirements. SAS Viya usually just starts a compute session in the back end. Kubernetes just blindly sees that there is a job coming in, but it doesn't know what it is. Instead, using queues associated with contexts, we can give that capability to the SAS administrator. And that’s, in the end, one of the big powers that SAS Grid Manager gave to SAS 9 administrators. A third reason is that with native Kubernetes if a job is preempted, it just goes away. It's terminated. There are no restart capabilities there. You get that restarting capability with SAS Workload Management.

What should I keep in mind when transitioning from SAS Grid Manager on SAS 9 to SAS Workload Management on SAS Viya? 

In the resources associated with this session, you'll see a migration link and documentation. I think the biggest thing you need to consider when you're migrating from SAS 9 to SAS Viya is your actual jobs that you're migrating. Once you know that those jobs are going to successfully run and in the SAS Viya world, then you want to go and re-architect from a grid perspective. How do I want those queues to be laid out? Do I want them to be the same or different in a cloud world? That way you can factor in differences between your virtual machines in the cloud versus your on-premise environment, which is a fixed environment.

SAS 9 Grid manager with LSF on SAS Grid platform enables us to install SAS on a high performing shared Filesystem. Does SAS Viya Workload Management also require a Shared filesystem? If so, what is the suggested filesystem?

SAS Workload Management on SAS Viya can operate with different file system and storage options, such as NFS or those provided by public cloud providers such as Azure, AWS etc., in addition to those considered more performant, such as GPFS. See Margaret Crevar’s paper https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2020/4312-2020.pdf

Since SAS Viya is available only with a cloud solution, not possible to install with on-premise machines like SAS 9, banking clients are still hesitant to move to the cloud because of security. Do you expect SAS Viya to have an on-premise option?

To clarify, SAS Viya operates on cloud-native principles, but has evolved from initial support on Azure, and is now supported on multiple cloud platforms (incl. AWS and GCP, in addition to Azure), and on-premise OpenShift K8s on VMWare. Plans are in place to support other platforms.

What is the alternative for SAS Enterprise Miner and SAS DI Studio in SAS Viya?

For capabilities earlier handled on Enterprise Miner (machine learning and advanced analytics), we offer Visual Data Mining and Machine Learning on SAS Viya. SAS Studio Flows and supported data management capabilities on SAS Viya help address some of the tasks earlier handled by SAS Data Integration Studio.

What is the license cost comparison between SAS 9.4 and SAS Viya?

Please reach out to your SAS AE to address this question.

Is it possible to preassign libraries to the SAS users in SAS Viya like we assign them in SAS 9 through Data library manager?

Yes, the same can be allocated and managed at the time of deployment or post-install.

Is there any concept of Authentication Domain, connection object in SAS Viya like we have in SAS 9?

This post, touching upon Auth Domains in the context of database access, may be a good starting point: https://blogs.sas.com/content/sgf/2017/08/17/sas-viya-sharing-credentials-for-database-access/#:~:te....

What happens to the end execution when one of the nodes for that node is down, on that part of the scheduled workflow?

Kubernetes handles the allocation of workloads to specific nodes. Workload Management does however have capabilities to allow for the preemption of workloads with options to restart the job or if checkpoint restarting is available for that job, to restart from a checkpoint.

SAS 9 has workspace server - used by SAS Enterprise Guide, SAS Pooled Workspace server used by SAS Web report studio, SAS Batch server used by SAS Batch process. Likewise, what are all the thin clients like SAS Studio, SAS Stored Process, SAS Web Report Studio available in SAS Viya?

Yes, SAS Viya offers web-based interfaces in keeping with overall objectives of low footprint, agile and accessible platforms. A mapping of these associations can be discussed with your AE.

Does a SAS Programmer have to learn anything special to work in SAS Viya programming?

SAS Data Steps working on compute workloads operate seamlessly in SAS Viya. There is a Content Assessment tool which deals with analyzing current SAS 9 codebases and highlighting areas where refactoring is suggested. There are additional enhancements with Cloud Analytics Services in SAS Viya through actions, which can be programmed with different interfaces.

Is there any way we can reuse SAS Enterprise Guide projects created in SAS 9 in SAS Viya using SAS Studio?

Please consult this article: https://communities.sas.com/t5/SAS-Communities-Library/Go-with-the-flow-migrating-Enterprise-Guide-P...

Shall we install SAS Viya 4.0 directly in our site or do we have to install SAS Viya 3.5 standard version and then apply hotfix or something like that to make it SAS Viya 4.0?

No, one can directly install SAS Viya 4.

Recommended Resources

SAS Workload Management

SAS Workload Management Blog

SAS Workload Management Documentation

The Evolution of SAS Workload Management

SAS Grid Manager Modernization: Do You Have a Plan?

SAS Workload Management on SAS Viya: Deployment Architecture

SAS Grid and SAS Viya: together to provide advanced workload management

Full-System Migration and Content Migration

Want more tips? Be sure to subscribe to the Ask the Expert board to receive follow up Q&A, slides and recordings from other SAS Ask the Expert webinars.