FSISUS04: How do you prioritize business critical functions over non-critical functions?
Determine what is defined as a business-critical process and workload, and protect and prioritize it. Model and prioritize individual functions and workloads by recording relevant metadata, such as interdependencies, SLAs for particular flows, and nuances of user access.
FSISUS04-BP01 Actively manage each business function and the allocation and configuration of resources
Prescriptive guidance
-
Use Amazon ECS Spot
compute for non-critical workloads such as end-of-month reconciliations. -
Use Amazon EC2 Dedicated Hosts
queues for priority jobs such as order initiation. -
Use Amazon ECR Lifecycle Policies for ephemeral ETL data such as ingestion ledgers.
-
Develop architecture strategies that use built-in queueing and buffering to offload non-critical tasks.
FSISUS04-BP02 FSI workloads serve the highest common denominator of application demands
Systems in financial services are built to serve the highest level of performance for retention, availability, and integrity. This leads to workloads that often exceed performance expectations or might not be respectful of ancillary or critical jobs and workflows. Breaking down a system into its component parts allows for a more fine-grained view of resource consumption and the trade-offs possible to balance SLAs against your sustainability goals.
Prescriptive guidance
Provide prioritization advice to customers on the following topics:
-
Prioritize at the organizational level: Determine what is defined as a business-critical process and workload and protect and prioritize it.
-
Prioritize at the SCP or OU level: Restrict AWS usage-based metrics on your Organizational Units' (OU) profiles and requirements. Batch-running processes that have extended SLAs can have dedicated accounts and permissions to restrict and reduce their carbon impact; for
example, select serverless preferences, choose specific instance types, or operate during specific processing hours. Development and test instances should have enforced central guardrails to limit Amazon EBS attachments or automatically pause and resume resources as needed.
-
Prioritize at the account level: Model and prioritize individual functions and workloads by recording relevant metadata, such as interdependencies, SLAs for particular flows, and nuances of user access. For example, investigations and warm access commonly take longer at a bank than its typical 35-day retention period.
-
Prioritize at the resource or tag level: Use tags to group and aggregate the management and reporting of resources. You may only have one critical flow but you likely monitor dozens of processes and receive millions of Event Notifications. Create a prioritization schema to determine which process matter most to your workload operations.
-
Prioritize at the job or object level: Not all jobs are born equal. Use mechanisms such as graceful termination of non-critical jobs and active workload management to help you prioritize at the job and object levels.
-
Prioritize resource allocation for critical generative AI applications in financial services: Implement right-sized generative AI models for different business criticality levels - use smaller, efficient models for non-critical functions. Evaluate if generative AI is necessary or if simpler approaches can achieve the same outcome. FSISUS05: How do you define, review, and optimize network access patterns for sustainability?
Assess and optimize network access patterns for sustainability. Pay attention to redundant layers and redirects or patterns generating excessive and unnecessary data movement.
