Backlog grew through community word of mouth Creating local spaces, sharing passion globally We will keep contributing to the community and the world, centered on our home of Fukuoka 06
Whether anyone watched costs depended on the team Learning Learned the framework Cloud FinOps (book), Framework, Certification Community Gave me know-how and passion FinOps X, Japan FinOps Meetup I want people working on FinOps in Fukuoka to learn from practice close to home 07
of tech Turning spend into value Shared data, timely decisions Financial accountability Engineering Those who spend can explain the cost The goal of FinOps is not cost cutting but maximizing the business value of technology 09
drives technology decisions 3 Everyone takes ownership for their technology usage 4 FinOps data should be accessible, timely, and accurate 5 FinOps should be enabled centrally 6 Take advantage of the variable cost model of the cloud The FinOps Framework guides practice with six principles 10
2 Business value drives technology decisions 3 Everyone takes ownership for their technology usage 4 FinOps data should be accessible, timely, and accurate 5 FinOps should be enabled centrally 6 Take advantage of the variable cost model of the cloud Per the principles, cost optimization is the job of cloud users, and a central team supports them 12
Finance Business Finance No cross-cutting role Business Costs relied on personal goodwill FinOps team (part-time) Engineering Engineering Teams Teams Teams Teams Teams Teams Teams Teams We aimed for a cross-cutting role that connects finance, business, and engineering 13
(part-time, virtual) Complex setups hide causes Probe Unfamiliar setup and history Impact checks need expertise Assess Not enough expertise on impact Hard to change what runs stably Change No change rights; proposals only Time Part-time; main job comes first Deadlines, security come first We could see the issues, but executing on them was too heavy a load 14
enabled centrally Owning teams FinOps team Own their costs Support with data and systems Investigate causes Investigate causes Prepare proposals Prepare proposals Verify effect and impact Verify effect and impact Approve Implement Implement (in part) Teams keep the decisions, and the FinOps team takes on investigating and verifying 15
Expertise required 2 Expert know-how across data via AI Stability blocks change 3 Verify effect and impact Case 1: Cost spike cause Case 4: Cost alert triage Case 6: Unused resources Case 2: Beyond Graviton Case 5: Commitments Case 7: Verify estimates Case 3: Budgets in ops 4 Foundation Data hub Tags & governance Safe permissions The FinOps team investigates thoroughly and drives execution with minimal burden on teams 16
2 Resources Identified the spiking account and resource 3 Metrics Total requests flat; only transfer volume spiked 4 Logs & traces By API, only the notice API spiked 5 Conclusion A terms update notice brought in users who rarely visit. Their browsers had no cache and fetched everything. We traced from billing to metrics, logs, and traces to find the cause 18
Graviton What we did CPU usage reduction after migration Migrated EKS nodes to Graviton (Arm) A‒D −46% to −60% E, F −15% to −33% G −1% barely reduced Results Total CPU usage −15% Response time −15% CPU usage breakdown by service A‒F G G accounts 2 of the total for about 3 For service G, the largest CPU user, migration alone barely reduced usage 19
CPU-heavy processing CPU time by API Inside the heaviest API Cause: SHA-512 acceleration (UseSHA512Intrinsics) was disabled at the time The profiler showed most CPU time went to hash computation in one API 20
measure Enabled JVM SHA-512 acceleration: -XX:+UseSHA512Intrinsics API response time Pod CPU usage EC2 cost (list, daily) Response time Pod CPU usage Node count EC2 cost −67% −45% −37% −30% Dig all the way to the cause and you can break the iron triangle 21
visibility and sharing Engineers' daily ops Per dept. & product Budget vs. actual vs. forecast Spot overruns early Month start: over Month end: on track Before Now Monthly review Daily forecasts Act before the gap grows and land within budget 22
AI reports first Cost alert Cost alert Human Agent Instruct Results Agent What grew and why Results Human Check intent AI investigates first, and people just confirm it is intended 24
Example purchase policy Org-wide and per account 1-year terms 3-year adds complexity EC2 / Fargate: Compute Savings Plans Discounts survive changes RDS / ElastiCache: buy RIs Confirm use; time upgrades to expiry Teams check plans on renewal Mind migrations and upgrades See the whole and each account on one dashboard Understand status with shared criteria and find teams that need support 25
Agent investigates Team decides and buys Contract expiry Policy, usage, contracts Check against plans Uncovered usage Draft a plan Buy, change, or skip FinOps Monitoring Proposal → purchase: Improve policy and agent 14 accounts, 9 were uncovered (lapsed / never) Each team decides on purchases from the proposals they receive 26
FinOps checks Team approves Candidates and costs Check priority, validity Check plans, approve Agent traces owners Propose to the team Delete after approval Run by FinOps or team Accounts involved Unattached disks Log groups NAT Gateway 6 9 109 7 FinOps sets criteria for unused resources and drives deletion through a system 28
of changes Change: new log delivery path Estimate Before Emit Store Send ✕ Monitor After Emit ✕ Relay Store Send Monitor Lower costs on a new path No more peak-time loss Actual Costs fell as estimated Zero delivery loss at peak Spikes caused log loss Decision Suppress unneeded logs Because we keep measuring before and after, we confirm effects and handle regressions quickly 29
by measurement Ex. 1: A renewed contract Ex. 2: A service we skipped Estimate Pricing and inquiries Actual Far above prior commit Decision Right-sized before renewal Actual Unexpected usage fees early in adoption Decision Skipped; revisit with data Keep measuring usage changes and reflect them in contract and adoption decisions 30
Tag policy Traces Platform Tag resources See, analyze, monitor Govern & fix tags Request fixes Investigate Approval Delete unneeded resources Access Read: all accounts Write: delete resources (limited) The cross-cutting foundation lets FinOps lead from investigation to execution, quickly and reliably 32
context Results Discount purchase Buy or renew from usage and contracts DBs to retire, envs to delete 2 teams bought with some items excluded CI transfer costs Move CI images to free-transfer storage Standard tooling builds it; no fetch Stopped fetching: no transfer cost, faster CI After a final check with the owning team, each initiative takes its best form 35
Ask teams to investigate FinOps investigates; teams just confirm Scope of support Platforms and issues Support from proposal to execution Culture Outreach and workshops Grow it by doing Support everything from building foundations to execution, and grow the culture through steady action 36
3 Verify effect across data via AI Connect billing data and telemetry Write down policy; automate proposals Go past recommendations; check effect and impact Trace billing changes down to specific code People can focus on checking and deciding Execute safely and respond quickly 4 Foundation Data hub Tagging Safe IAM and impact Cross walls, dig deep, deliver The FinOps team investigates thoroughly and drives execution with minimal burden on teams 37
nulabinc
holman
axbom
erikaheidi
mkilby
pedronauck
ivargrimstad
aleyda
geoffreycrofte
skipperchong
gurzu
codingconduct
palkan
