faisal mushtaq - an enterprise cloud cost management framework
- 1. Cost Management in the Cloud for Enterprises Faisal Mushtaq, GlobalLogic
- 2. About the speaker • 20+years of experience in building global organizations and innovativeproducts • VP Technology & Head of Global Cloud Practice,, GlobalLogic • Key role in 3 startups with $200+Million exit • VP Engineering PLUMgrid (Acquired by VMware) • VP Mobile Video ProductLine (Allot Communications) • VP Engineering Ortiva Wireless (Acquired by Allot Communications) • Founding Member Wichorus (Acquired by Tellabs) • 10+years at Cisco Systems • Extensive experience in Cloud technologies OpenStack, Containers, SDN, GCP, AWS, Networking
- 3. Agenda • Cost Management Challenges • A Cost Management Framework • Automation Tools Landscape • Case Study • Q&A
- 5. Why is Cost Management Cloud sprawl important in the cloud? Bill shocks Unmanaged costs often lead enterprises to abandon cloud journey Dev/Test costs often exceed budgets Cloud costs manifestdifferently (CAPEX vs OPEX) Up to 6X of production
- 6. Optimization • Apps not utilizing cloud features to optimize cost (e.g. autoscaling) • Cloud platform discounts not utilized • Periodic reviews not in place for validating evolving application architectures Ownership • Who owns the cloud resources? • Who pays for them? Budgets and TCO • Has sizing and TCO been done? • Have Pre-Go-Live costs been considered? • CSP provided credits are a double edged sword Visibility • No visibility or insights for stakeholders • Lack of continuous cost monitoring • Cost projections not available → No proactive response Governance • Governance systems not in place, or not enforced • Who gets to create what resources? Key Cloud Cost Management Challenges Automation • Governance not enforced through tools and automation • Non-existent or manual cost and compliance reporting
- 7. A Cost Management Framework
- 8. Approach Think about costs from Day 1 2 3 4 5 6 Enable behavioural change for costs in cloud Move from CAPEX to OPEX mindset Think about cost impact of all design decisions Implement, enforce and automate governance mechanisms Do TCO for before deploying 1
- 9. A Cost Management Framework Automation Governance Initial Planning (Sizing and Provisioning) Operational visibility and forecasting Cost optimization
- 10. Governance
- 11. Transform provisioning practices for the cloud De-centralized – App teams can provision resources, but IT team tracks usage Chargeback/Showback – IT teams charge app teams for usage Build reference architectures and templates (EA responsibility) Automate provisioning of reference architectures and templates Use a cloud aware CMDB tool (e.g. ServiceNow) Cloud Resource Ownership
- 12. lT teams can define the governance practices Governance Application teams implement practices for their environment IT teams track conformance to governance practices • Labeling • Resourcelifecycle 1 2 3 Cloud Resource Provisioning
- 13. • Rule of thumb is to use tags for resource management and labels for resource identification, grouping, searches & billing • Define list of labels and tags to be applied Identification/Classification Tags • BU/Cost Center • Application • Owner-email– application owner/group • Environment– Prod/Dev/Test/QA/Perf • Environment-Name– Prod1a, Dev4 etc. • Chargeback/ShowbackID • created-by - User who created resource. • role = <db, appserver, proxy, etc.>- Classify by application role Tagging and Labelling
- 14. Build or use a lightweightinventorymanagement system • Continuously track cloud sprawl • Report current inventory, new resources, projected costs • Do periodic gap analysis Inventory Management
- 15. Initial Planning
- 16. TCO and Budgeting • Use max CPU/RAM for budgeting • Do initial right-sizing based on on-prem monitoring data • For Dev/Test, consider uptime hours (9x5 as opposed to 24x7) for TCO calculations • Enforce budgets from Day 1 Service Catalogsand Provisioning • IAM policies to restrict creation of only allowed services • Build IT certified base images and templates for reference architectures • Publish and enable self-provisioning through tools like ServiceNow • Integrate with approval processes Sizing and Provisioning
- 17. Operational Visibility and Forecasting
- 18. 1 23 Daily Reporting including cost, utilization, non-conformantresources Analytics based recommendationsMonthly billing reports to required stakeholders Reporting Approach
- 19. Reporting Recommendations Automation Exceptions Billing Optimization Inventory ResourceOptimizationTagging Cloud Accounts Billing Data Audit Data API Utilization Data Reporting and Automation Architecture
- 20. Data Points Cost (by app/environment) • Daily, MTD and Projected Monthly Spend • Budgeted vs Actual, and Overrun Projection • Alerts on any changein usage pattern and/or budget overruns Utilization • Show unused resources - Unattacheddisks - Orphanedsnapshots - UnallocatedIPs - Unused/unaccessed Storage(recommendarchival) • Show underutilized resources - Showinstancesandenvironmentswithnoutilization Inventory • CurrentInventory • New resources created + corresponding cost • New projected monthly spend based on new resources Conformance • List of resources withouttags and labels • List of resources notconfirming to naming conventions • List of instances based on older versions of baseline images to Report
- 21. • Rightsizing • Eligible Reservation Planning/Committed Use - Results in up to 24%-57% potential savings • Instance Scheduling • Eligible Spot/Pre-emptible Instances - Results in up to 60-80% potential savings • Reserved/Committed Instance Renewal Alerts (for instances with approaching expiry dates) Recommendations (with savings projections)
- 22. Cost Optimization and Automation
- 23. • Effective Cost Optimization is a combination of two approaches: - In-depth operational visibility for action by app teams - Automatic resource cleanup (as a fall-back) • Approach - Stakeholders given opportunities to take action, or register exceptions - No action from stakeholders recommendations are automatically actioned - The two approaches, used in combination will lead to better awareness and accountability Cost Optimization Approach
- 24. Automation Approach • If automation not available in tools, build in-house. • Start small and grow the automation catalog. • No one tool will solve all cost management problems. Build and integrate tools as services (see Reporting and Automation Architecture)
- 25. Infrastructure Optimization • Instance Rightsizing - Right-size individual instances based on utilization • Cleanup of unused resources - As per utilization reports • Cleanup of underutilized resources - Minimal or No CPU utilization - Minimal or No Disk activity - Minimal or No IO activity • Instance scheduling - Based on spikes in usages patterns - Instance scheduling for dev/test servers that don't need to be run 24/7 • Instance modernization - Based on new instance families released by the cloud providers (often cheaper) • Cleanup of other Cloud Services - Based on Cloud based metrics
- 26. Billing Optimization • Production - Use reserved/committed use discounts - Start with – 30%-40% servers (before stabilization = immediate cost savings) - End with – 100% servers (after app stabilizes in cloud) - Use Cloud Services to get TCO of reference architectures before deployment • Dev/Test - Enable committed use & spot/preemptible instances - Start with – 10% servers (Eg., Toolchains – Gerrit/Jenkins etc.) - End with – 100% (after app stabilizes in cloud) - Spot/Preemptible instances for environments that can be torn down and recreated - Integrate use of spot/preemptible instances with DevOps build processes
- 27. • Restrict number of Dev/Test environments - Review DevOps pipelines - Evaluate alternate and efficient ways for developers to unit test their code - Right-size Dev/Test environments • For Consultants/Vendors - Customer doesn’t mind spending NOT AN EXCUSE - Think about cost impacts for all design decisions DevOps Optimization
- 28. • Enforcing Labels - Automatically add certain missing tags such as created-by (use to track creators of orphaned resources) - Create and maintain virtual tags for Cloud services that don't support tags yet, in the inventory management system Other Automation Opportunities
- 30. Automation Tools Landscape Concern AWS Azure GCP Third Party/Custom Initial Sizing • AWS Cost calculator • AWS TCO Calculator • AWS Applicationdiscovery service • Azure Pricing Calculator • Azure TCO Calculator • GCP Pricing Calculator • GCP TCO Calculator • GL’s custom tools Operational Visibility and Forecasting • TrustedAdvisor • AWS Organizations • Tags • Azure Advisor • Azure Cost Management • GCP Labels • GCP Projects • GCP Data Studio • Cloudability • Cmpute.io • Cloudcheckr • Cloudyn • Botmetric • RightScale Cost Optimization • TrustedAdvisor • AWS Lambda • Azure Automation • Azure Cost Management • Google CloudFunctions • Cloudability • Cmpute.io • Cloudcheckr • Cloudyn • Botmetric • RightScale • Monitoring tools (Nagios, DataDog, Sensu) • Skeddly, Ylastic, etc.
- 31. Case Study A Retail Customer Migrating to GCP
- 32. • Migrating their entire landscape to GCP over a 2 year period • About 40% applications already migrated to GCP • Initial sizing and planning done • Manual Cost Reporting (automation in progress) Current State
- 33. Challenges • No upfront cost management strategy and governance framework - Some governance mechanisms defined but not enforced - Other mechanisms being implemented on a 3-6 month timeline • Cost management looked at after the fact • No granular visibility into spends • Actual cloud spends substantially larger than budgets - Monthly spends – Actual 5x times the budgeted spent - GCP discounts not utilized (committed and pre-emptible instance discounts) - No Chargeback - App teams create resources, IT team pays for them • Dev/Test spends substantially larger than Prod - ~40/50 Dev environments for some apps
- 34. • Immediately start using a Cost Management Tool - Build reporting and recommendations - Build inventory management • Enforce labelling • Enforce clean-up of non-conformant resource • Use cloud discounts Recommendations Given