Primary Storage

IBM Storage Insights presents AI-Powered "Workload Placement Advisory" for IBM's FlashSystem Family !

Author: @Binayak Dutta

Background

For a storage planner, provisioning storage for newer applications (workloads) is one of important job function. And there are few pertinent questions -

1. Are my existing systems fully utilized ? Can I accommodate the new workload in existing fleet of systems? 
2. Which of the existing systems (if at all) will be best placed to service the new workload ?
3. For how long can new workload be accommodated (in existing fleet) ? And thus help defer additional cost of procurement.

Presently, answers to such questions (and hence the decisions) may vary depending on the administrator's expertise and in some cases a set of best practices and thumb rules. And often such decisions could be with lesser confidence, conservative and lack awareness of the recent state of different systems.

The new feature, Workload Placement Advisory is offered to help Storage Planner in this decision making process. The feature is backed by a patent-pending system and method for evaluating fitment of workload in a IBM Storage FlashSystem

Feature Introduction

Workload Placement Advisory as a feature will empower the user to model various scenarios (varying workload characteristics, plan durations, system search-space) and evaluate the fitment of existing systems. The systems will be ranked by their fitment and backed by details which explains the behaviour and influence of important system KPIs in the ranking process. 

Feature Walk-Through

The feature is accessible in Storage Insights through Legacy Dashboard(Insights features) or New Experience Dashboard (Planning features). Once invoked, the user lands in the feature page. 

Figure 1.

The navigation from input to output is as below:

1. User approximates various characteristics of the new workload. Two possible paths -
   1. To ease the experience, user is provided with option to choose from pre-defined workload templates (like OLAP, OLTP, Back up systems etc.) and t-shirt size (small, medium, large). Once a template is selected, user can fine tune the required properties.
   2. Alternatively, use can select "custom" and tune in all the properties.
2. Once the new workload characteristics are defined, user will move on to specify the planning duration that is the date in future (starting today) till which the new workload will need to be accommodated. With time, capacity and performance demands of many workloads tend to out grow the available resources. Hence it is advisable to put a realistic planning duration like a quarter, half-year or a year ahead.
3. Next step is to choose a subset of existing systems which user wished to be evaluated. This helps by
   1. filtering out systems which are out of bound (for e.g. development and test systems which cannot host production workload or vice versa)
   2. hand picking a preferred subset of system which are user or department is inclined to use
   3. Fitment evaluation process is compute heavy as it evaluates multiple KPIs. Hence restricting the list of systems helps reduce the wait time to reasonable limits.
4. Once the user inputs the workload characteristics, planning duration and prospect list of systems, user is ready to hit evaluate button which will triggers the fitment evaluation process. Following are some important steps 
   1. Based on make, model, system software version, configurations and current workload, performance benchmark for each system is derived. This includes the thresholds for various KPIs (metrics) which the system can sustain
   2. Historical capacity and performance trends of the system and new workload demands are leveraged to forecast and project the future growth
   3. A fitment score per KPI is assigned based on proximity of KPI projection with threshold limits
   4. A weighted average of scores is assigned to the system as its fitment evaluation score
   5. Systems are ranked by scores and surfaced
5. Users thus have a ranked list of systems evaluated for fitment against the new workload for the chosen planning duration. To gain understanding and confidence in the ranking, a detailed explainability feature is provided. On clicking any of the ranked systems, a set of graphs are displayed which explains how the KPIs are projected to behave in future with or without the new workload. The horizontal lines depicts the KPI specific threshold limits. Figure below (Figure 2) provides a snapshot of the explainability feature page.

Figure 2.

By changing various input criteria, user can perform scenario modelling (what-if analysis) and use this feature to answer different questions like - 

• What-If, the capacity growth-rate of new workload increases to 20%. Will it still fit-in the earlier recommended system ?
• Can the earlier recommended system sustain the new workload for an extended duration (say 2 years) ?
• Can I find a compatible system if the latency tolerance for new workload is set to 2ms (instead of 1ms) ?

Hopeful that the feature will be used to its full potential as we plan to extend the planning features further.

Thank you.

References:

1. For details on IBM Storage Insights, please refer our product page https://www.ibm.com/products/storage-insights
2. For a hands-on experience with IBM Storage Insights, please use our demonstration link https://demo.insights.ibm.com/
3. For a quick summary of features released with IBM Storage Insights in 2Q2024, please refer 

   https://www.ibm.com/docs/en/storage-insights?topic=whats-new

4. For a video demonstration of Workload Placement Advisory feature, please refer our youtube link IBM Storage Insights - Workload Placement Advice
5. For recent updates and demos across IBM Storage Product Portfolio, please refer our youtube channel @StorageGuru