stdsquare²
🎓 Kelas
stdsquare / materi / slides / pertemuan-13
Tema
Japan
Arcade
Dark Retro
Font
‹ Daftar slide Pertemuan 13: FinOps Cloud Cost Management (Tagging, RI, Spot, Optimization)
PEBD6027 · RPS minggu 14 · 2x50 menit

FinOps & Cloud Cost

Cost Visibility · Allocation · Optimization · Reserved Instance · Spot

Peta Pembelajaran Hari Ini

Sub-CPMK 13: mengelola dan mengoptimalkan biaya cloud. Empat indikator capaian.

Jam ke-1
  • FinOps framework & culture
  • Cost visibility: tagging, billing
  • Cost allocation per tim/business unit
Jam ke-2
  • Cost optimization: RI, SP, Spot
  • Right-sizing & idle removal
  • Studi kasus: hemat 30-50% bill cloud
Posisi 14P: setelah DevOps P12, hari ini FinOps. P14 studi kasus Indonesia.

Mengapa FinOps Penting?

Cloud pay-as-you-go = mudah beli, mudah boros. Tanpa FinOps, cost runaway.

Cloud Waste
~32%
Flexera 2023 State of Cloud — % cost terbuang
Hemat via FinOps
30-50%
Studi FinOps Foundation 2023 — saving rata-rata klien
Startup Indonesia sering kebangkrutan karena cloud bill meledak. Tanpa tagging & alert, tidak sadar sampai terlambat.
Bagian 1 · 1/4
FinOps Framework
Diskusi: siapa yang harus own cloud cost — engineering, finance, atau business unit?

FinOps Foundation Framework

FinOps = budaya + praktik + tool. 6 principle + 3 phase lifecycle.

6 Principles
  • Collaboration: tim + finance + business
  • Business value everyone responsible
  • Single source truth data
  • Prioritize cost vs usage visibility
3 Phase Lifecycle
  • Inform: visibility, allocation
  • Optimize: right-size, RI, Spot
  • Operate: automation, continuous
Finops maturity: Crawl (visibility) → Walk (optimize) → Run (automate). Sebagian besar perusahaan RI di Crawl.

Cost Allocation via Tagging

Tagging = fondasi FinOps. Setiap resource harus dapat ditelusuri ke business unit.

Tag Strategy Wajib
  • Environment: dev/staging/prod
  • Team: backend, frontend, data
  • Project / Product
  • Cost Center / Business Unit
Best Practice
  • Tag policy via SCP
  • Tag saat creation, bukan retrospective
  • Report untagged resource
  • Cost allocation tag activate
Tanpa tagging: tagih cloud = tagih listrik tanpa alamat. Tidak bisa allocate per tim. Foundation wajib FinOps.

Studi Kasus: Bank Jago FinOps Team

Bank Jago punya tim FinOps dedicated untuk control cloud cost multi-account.

Public case study Bank Jago FinOps (dihedge): AWS Organizations multi-account, tagging policy wajib via SCP, AWS Budget alert per business unit, Reserved Instance + Savings Plan strategis, Spot Instance untuk workload batch, auto-terminate dev/test di malam hari via Lambda, monthly FinOps review cross-functional. Hasil: hemat 35-45% bill cloud setelah 6 bulan.
Praktik Konkret
  • Tag 100% resource
  • Budget alert per tim 80% & 100%
  • RI 1-year untuk steady workload
  • Spot untuk dev/test/batch
Hasil
  • Bill visibility per BU
  • Anomaly detection cepat
  • 35-45% hemat vs pre-FinOps
  • Engineer awareness cost
Pelajaran Bisdig: FinOps bukan IT-only. Cross-functional team: Eng + Finance + Business Unit leader.

Hitung dari Nol: Right-Sizing EC2

Startup EC2 10x m5.2xlarge (8 vCPU 32GB), utilization rata-rata 15%. Berapa hemat?

LangkahStatusNilai
110 × m5.2xlarge on-demand$3,52/jam × 24 × 30 = $2,534/bln
2Avg utilization 15%(Cuma 15% dari 8 vCPU = 1,2 vCPU used)
3Right-size ke m5.large (2 vCPU)Sesuai kebutuhan real
410 × m5.large on-demand$0,88/jam × 24 × 30 = $634/bln
5Hemat right-sizing$1,900/bln ($22,800/tahun)
6Plus: RI 1-year no upfront (40% off)Hemat tambahan $253/bln
7Total hemat$2,153/bln ($25,800/tahun, 85% hemat)
Right-sizing paling impactful. Monitor via CloudWatch metric 14 hari sebelum downgrade. Bisa upgrade lagi bila perlu.
Bagian 2 · 2/4
Commit Discount: RI, SP, Spot
Diskusi: bila workload steady 24/7 untuk 3 tahun ke depan, pilih RI atau Savings Plan?

Reserved Instance (RI)

RI = commit 1 atau 3 tahun untuk discount. Cocok steady workload.

Tipe RIDiscountFlexibility
Standard RI75% (max)Locked ke instance type + AZ
Convertible RI60%Bisa swap instance type
Scheduled RIVariableBeli window waktu tertentu
1-year No Upfront40%Bayar per bulan, no commit awal
1-year All Upfront50%Bayar penuh di awal
3-year All Upfront75%Bayar penuh, hemat max
RI 3-year hemat max tapi lock-in. Bila tech berubah, rugi. Untuk startup volatile: 1-year No Upfront aman.

Compute Savings Plan (CSP)

CSP = commit dollar per jam, lebih fleksibel dari RI. Diskon apply otomatis.

Tipe CSP
  • Compute SP: any instance, any region
  • EC2 Instance SP: specific instance family
  • SageMaker SP: ML workload
Manfaat
  • Discount 40-66% seperti RI
  • Auto-apply ke highest discount
  • Swap instance anytime
  • Cover Fargate & Lambda juga
CSP lebih fleksibel dari RI. Untuk sebagian besar kasus, Compute SP adalah pilihan modern.

Spot Instance untuk Batch

Spot = idle capacity AWS dengan discount 90%. Risiko: termination 2 menit notice.

Karakteristik
  • Discount 90% vs on-demand
  • Termination notice 2 menit
  • Reclaim bisa kapan saja
  • Spot Block (deprecated, pakai fleet)
Use Case Cocok
  • Batch processing
  • ML training
  • CI/CD worker
  • Image processing
Tidak Cocok
  • DB primary
  • Web server latency-sensitive
  • Stateful service
  • Long-running transaction
Mitigasi Risk
  • Checkpoint progress ke S3
  • Spot Fleet multi-AZ
  • Auto-recover via ASG
  • Kubernetes Spot Node + interruption handler
Spot hemat 90% tapi butuh architect untuk fault-tolerant. Bukan untuk semua workload.
Bagian 3 · 3/4
Cost Optimization Strategy
Diskusi: bila tagihan AWS Rp 500 juta/bulan, apa langkah optimasi pertama?

Visibility Tools: Cost Explorer & Trusted Advisor

Cost Explorer = dashboard cost. Trusted Advisor = rekomendasi optimasi. Wajib pakai.

AWS Cost Explorer
  • Cost per service, tag, region
  • Forecast 12 bulan ke depan
  • RIs/SPs coverage report
  • Anomaly detection
Trusted Advisor
  • Idle EC2 check
  • Low utilization EC2
  • Unassociated EIP
  • S3 bucket versioning lama
Trusted Advisor free tier: 7 check. Business/Enterprise support: 100+ check. Untuk startup, free tier cukup mulai.

Cost Optimization Checklist

Checklist optimasi cost berdasarkan FinOps Foundation. Target 30-50% hemat.

KategoriActionHemat Estimasi
Right-sizingDowngrade EC2 underutilized20-40% compute
RI/CSP CommitBeli 1-year No Upforward steady30-40% compute
Spot InstanceMigrate batch & tolerant fault60-90% compute
S3 LifecycleTier Glacier untuk archive30-60% storage
Egress OptimizeVPC Endpoint S3/DynamoDB30-50% network
Idle ResourceTerminate dev/test malam10-20% total
Multi-AZ ReviewMulti-AZ only for HA-critical20-40% compute
Log LifecycleCloudWatch Logs → S3 Glacier30-50% log
Eksekusi prioritas: right-sizing + RI/CSP + Spot = 60% hemat. Storage & network = 30% hemat tambahan.

Studi Kasus: Tokopedia Spot untuk ML

Tokopedia pakai Spot Instance untuk ML training pipeline, hemat jutaan dolar per tahun.

Public case study AWS Tokopedia ML (dihedge): ML training pipeline SageMaker + EC2 Spot Fleet. Sebelumnya on-demand 24/7 sangat mahal untuk training intermittent. Migrasi ke Spot: checkpoint ke S3 setiap epoch, auto-recover via Spot Fleet, fallback ke on-demand bila no Spot capacity. Hasil (dihedge): hemat ~70-80% cost ML compute, throughput training same, cost per experiment turun 5x.
Arsitektur
  • SageMaker Training Job + Spot
  • Checkpoint S3 setiap epoch
  • Spot Fleet multi-AZ
  • On-demand fallback
Hasil Bisdig
  • Hemat 70-80% ML compute
  • Throughput training same
  • Cost per experiment turun 5x
  • Lebih banyak eksperimen ML
Pelajaran: Spot bukan hemat murah, tapi enable business case baru yang sebelumnya tidak feasible.
Bagian 4 · 4/4
Automation & Penutup
Diskusi: bagaimana cegah developer spin up resource liar tanpa approval?

Automation FinOps

FinOps maturity Run: automation built-in. Bukan retrospective manual review.

Auto-Terminate
  • Lambda cron stop dev/test malam
  • ASG scale-to-zero weekend
  • Auto-delete untagged resource after 7 hari
Policy-as-Code
  • SCP block expensive instance
  • Tag policy wajib
  • AWS Config rule alert
  • Budget action via Lambda
Auto-terminate dev/test malam: hemat 60% cost env non-prod. Risiko: engineer butuh akses malam? Pakai SSM start-stop on-demand.

FinOps Culture & Cross-Functional

FinOps bukan IT-only. Cross-functional team: Engineering + Finance + Business.

Engineering Role
  • Architecture cost-aware
  • Tag resource saat create
  • Right-size via metric
  • Use Spot/RI saat cocok
Finance Role
  • Budget & forecast
  • Variance analysis
  • Cost allocation per BU
  • Procurement commit RI/CSP
Business Role
  • ROI cloud investment
  • Prioritize workload cost vs value
  • Make vs buy decision
  • Cost per customer metric
FinOps culture = engineering paham cost, finance paham teknologi, business paham ROI. Budaya shared responsibility.

Widget: TCO Cloud vs On-Premise (Reuse)

Masukkan jumlah server on-prem, biaya CapEx awal, biaya listrik pendingin ruangan, dan harga instance cloud. Lihat breakeven dan total cost of ownership 3 tahun. Sudah dipakai matkul lain, sekarang Anda pakai untuk justifikasi FinOps.

Ringkasan Kunci Pertemuan 13

Hari ini Anda menguasai FinOps: visibility, allocation, optimization, automation.

Visibility
Cost Explorer + tagging + budget alert. Foundation wajib FinOps.
Optimization
Right-sizing + RI/CSP + Spot. Target hemat 30-50%.
Culture
Cross-functional Eng + Finance + Business. Shared responsibility.

Tiga lensa P13: tagging sebagai fondasi · commit discount untuk steady · Spot untuk fault-tolerant.

Persiapan P14: Studi kasus Indonesia (Gojek, Tokopedia, Bank Jago). Review seluruh semester. Kuis 5 menit soal RI vs Spot.