🏭 Infrastructure Evolution — Dari Mainframe ke AI-Native Edge Cloud
Infrastruktur bukan hanya tentang "server fisik vs cloud" — ia adalah perjalanan 8 lapisan dari 1960s bare-metal mainframe hingga 2026 AI-native edge cloud . Vault punya catatan untuk cloud-security, container/k8s, ansible-hardening tapi tidak ada peta evolusi yang menyatukan semuanya. Catatan ini memetakan evolusi datacenter, virtualisasi, cloud, container, serverless, edge, sampai AI-infrastructure-ops, dengan trade-off matrix dan decision framework.
Daftar Isi
1. Premise — Evolusi Infrastruktur Itu Spiral 2. Eight-Layer Infrastructure Evolution 3. Layer 0 — Bare-Metal & Mainframe) 4. Layer 1 — Virtualization (Hypervisor) 5. Layer 2 — IaaS Public Cloud 6. Layer 3 — PaaS & Container Orchestration 7. Layer 4 — Serverless & FaaS 8. Layer 5 — Edge Compute
1. Premise — Evolusi Infrastruktur Itu Spiral
Bukan linear — infrastruktur ulang-alik antara centralization ↔ decentralization , control ↔ abstraction , specialization ↔ generalization :
1960s: Centralized mainframe ↔
1970s: Centralized mainframe ↔
1980s: Decentralized PC server ↔
1990s: Centralized 3-tier ↔
2000s: Decentralized SOA ↔
2010s: Centralized cloud ↔
2015s: Decentralized containers ↔
2018s: Centralized orchestration ↔
2020s: Decentralized edge + multi-cloud ↔
2026: AI-orchestrated hybrid ↔
Prinsip penting: setiap lapisan memecahkan masalah lapisan sebelumnya, sambil menciptakan kategori masalah baru yang akan melahirkan lapisan berikut.
2. Eight-Layer Infrastructure Evolution
Layer Era Karakteristik Masalah yang Dipecahkan Masalah Baru 0 1960s+ Bare-metal - Utilization rendah (2-5%) 1 2000s Hypervisor Utilization naik ke 60-80% VM sprawl, license cost 2 2008+ IaaS Cloud CapEx → OpEx Vendor lock-in, cloud bill shock 3 2014+ Container + K8s Density lebih, faster deploy Operational complexity 4 2015+ Serverless / FaaS Zero ops Cold start, lock-in 5 2020+ Edge Compute Sub-50ms latency Distributed observability 6 2024+ Multi/hybrid cloud Vendor freedom Network complexity 7 2026+ AI-Orchestrated Inf.rastructure Self-healing Hallucinated automation
3. Layer 0 — Bare-Metal & Mainframe
3.1 Karakteristik
┌───────────────────┐
│ Single physical │
│ server + OS │
│ ├─ Apps diinstall │
│ └─ Services run │
└───────────────────┘
Pro:
Total control
Performance predictable
No virtualization overhead
Con:
Utilization 5-15%
Provisioning manual (minggu)
Single point of failure
Hardware lock-in (5-7 year refresh)
4. Layer 1 — Virtualization (Hypervisor)
4.1 Arsitektur
┌────────────────────────┐
│ Hypervisor (ESXi, KVM) │
├─────┬─────┬─────┬─────┤
│ VM │ VM │ VM │ VM │
│ Web │ DB │ App │ DC │
└─────┴─────┴─────┴─────┘
│ │ │ │
└───────┴───────┴───────┘
Single physical host
4.2 Evolusi Hypervisor
Tahun Perkembangan 2003 VMware ESX 2.0 (industry standard) 2005 Xen (open source) 2007 KVM releases (Linux native) 2010 Hyper-V, public cloud infra use KVM/Xen 2013 Docker popularized containers (LXC kernel) 2015 KVM+OVS standard open-source stack 2020 Unikernel, micro-VM (Firecracker)
4.3 Pro & Con
Pro Con Utilization naik ke 60-80% VM license cost Fast provisioning (menit) VM sprawl tanpa governance Snapshot + clone Hypervisor overhead (5-15%) Live migration ”Pet vs Cattle” anti-pattern
5. Layer 2 — IaaS Public Cloud
5.1 Arsitektur
┌────────────────────────────────────────────────────┐
│ Region (geographic area) │
│ ├─ AZ (Availability Zone) │
│ │ ├─ Compute (EC2/VM/GCE/Instance) │
│ │ ├─ Storage (S3/Blob/GCS/Object) │
│ │ ├─ Network (VPC/VNet/VPC) │
│ │ └─ Managed DB (RDS/CloudSQL/SQL) │
│ └─ AZ │
└────────────────────────────────────────────────────┘
5.2 Tiga Cloud Besar + Specialized
Tier Service Hyperscale (3 besar) AWS, Azure, GCP Secondary (alibaba, tencent, ibm, oracle) Banyak Sovereign (region-locked) Eropa, China, GovCloud Specialized (AI/ML focused) Lambda Labs, Paperspace, Coreweave Bare-metal cloud Equinix, Packet (now Equinix)
5.3 Pro & Con
Pro:
CapEx → OpEx (bayar pakai)
Global presence dalam menit
Managed services (RDS, Elasticache, dll)
Pay-per-use scaling
Con:
Vendor lock-in luas
Cloud bill sprawl (unpredictable)
Egress fees mahal
Compliance & data residency kompleks
6. Layer 3 — PaaS & Container Orchestration
6.1 Container vs VM
Aspect Container VM Boot time <1s ~30s Image size 100MB 10GB Density 50-100/host 10-20/host Isolation Process-level Hardware-level Use case Microservices, ephemeral Stateful, multi-tenant
6.2 Container Orchestrators
Tool Status Use Case Kubernetes De facto standard Most workloads Nomad HashiCorp alternative Simpler ops ECS (AWS)AWS native Tight AWS integration Cloud Run GCP managed Stateless containers Fly.io Edge-first Simple regional K3s Lightweight K8s Edge, IoT, dev
6.3 K8s Layer Stack
┌────────────────────────────────────────────┐
│ Workload (Pod, Deployment, StatefulSet) │
├────────────────────────────────────────────┤
│ Service Mesh (Istio, Linkerd, Cilium) │
│ Network Policy, mTLS │
├────────────────────────────────────────────┤
│ Runtime (containerd, CRI-O) │
├────────────────────────────────────────────┤
│ Compute (kubelet, scheduler, controller) │
├────────────────────────────────────────────┤
│ Network (CNI — Calico, Cilium, Flannel) │
├────────────────────────────────────────────┤
│ Storage (CSI, persistent volumes) │
├────────────────────────────────────────────┤
│ OS Layer (Linux + optionally Windows) │
├────────────────────────────────────────────┤
│ Hardware / Hypervisor / Bare metal │
└────────────────────────────────────────────┘
Koneksi ke Vault:
7. Layer 4 — Serverless & FaaS
7.1 Definisi
[Function code (Lambda/Function)]
↓
[Event trigger — S3, queue, schedule, HTTP]
↓
[Provider allocates runtime, runs, returns result]
↓
[Provider charges per ms + per invocation]
7.2 Serverless Spectrum
Service Type Control Cost Efficiency Use Case FaaS (Lambda) Minimum Highest (idle = 0) Event handler Container FaaS (Cloud Run) Medium High Stateless API Backend-as-a-Service Various High Mobile/web app SaaS replacements Zero Highest Non-differentiating workloads
7.3 Cold Start Problem
Mitigation Effect Provisioned Concurrency Eliminates cold Warm-up calls Mitigates partially SnapStart (Java) <100ms vs 5s WebAssembly <5ms cold Reserved instances Eliminates for hot paths
8. Layer 5 — Edge Compute
8.1 Arsitektur
┌─────────────────────┐
│ Cloud (origin) │
│ - Heavy processing │
│ - ML training │
│ - DB of record │
└─────────▲───────────┘
│
│
┌───────────────┼───────────────┐
↓ ↓ ↓
┌──────────┐ ┌──────────┐ ┌──────────┐
│ Edge POP │ │ Edge POP │ │ Edge POP │
│ ~10ms │ │ ~20ms │ │ ~15ms │
│ Sub-ms │ │ │ │ │
└──────────┘ └──────────┘ └──────────┘
8.2 Provider Edge Compute
Provider Product Latency Capacity Cloudflare Workers <5ms globally V8 Isolates / WASM Vercel Edge Functions Sub-50ms V8 + WASM Fastly Compute@Edge Sub-50ms WASM / Lucet AWS Lambda@Edge / CloudFront Functions 5-50ms Node.js/Python Akamai EdgeWorkers Sub-50ms V8 Fly.io Regional Machines <50ms Full containers
8.3 Use Case Pattern
Pattern Edge Component Cloud Component Authentication check Validate JWT, redirect Issue token, audit Image optimization Resize, format convert Store original A/B test routing Apply variant Analytics Real-time multiplayer Synchronize state Leaderboard persistence Geo-restriction Check IP-based access Authorization policy Live video sub-second caption Stream process Full ML model
Koneksi ke Vault:
9. Layer 6 — Multi/Hybrid Cloud
9.1 Karakteristik
┌────────────────────┐
│ Multi-cloud fabric │
│ Management plane │
└─────┬────────┬─────┘
↓ ↓
┌──────┐ ┌──────┐
│ AWS │ │ GCP │
└──────┘ └──────┘
+ Optional on-prem
9.2 Use Case
Vendor freedom — exit strategyBest-of-breed — pakai tiap cloud terbaikCompliance — data sovereignty per regionDR — multi-region = resilience
9.3 Trade-offs
Pro Con Vendor bargaining power Network complexity Avoid single-point-of-failure Cost monitoring across accounts Compliance per-region Operational expertise multi Best-of-breed per service Skill fragmentation
9.4 Multi-Cloud Stack (2026)
Layer Tools Control plane Crossplane, Terraform Cloud, Pulumi K8s federation Cluster API, KubeFed, Admiralty Service mesh cross-cloud Istio multi-primary, Cilium cluster mesh Storage sync Restic, Velero, MinIO multi-site Observability Grafana Cloud, Datadog, Honeycomb Cost management Vantage, CloudHealth, nOps
10. Layer 7 — AI-Orchestrated Infrastructure
10.1 Definisi
[User Intent] → [AI Agent]
├─ Plan
├─ Decide (which cloud, which service)
├─ Provision
├─ Configure
├─ Monitor
├─ Mitigate
└─ Optimize
10.2 Capabilities (2026 Realistic)
Capability Maturity Anomaly detection in metrics Production-ready Auto-scaling triggered by ML Production-ready Cost optimization recommendations Production-ready Natural language → IaC Emerging (errors common) Self-healing infra (restart service) Production-ready for known patterns Auto-architect (design new system from intent) R&D Self-modifying infra DANGEROUS — research only
10.3 Pattern AIOps
Pattern Contoh Predictive autoscaling ML predict traffic 1h ahead Anomaly-based alerting Statistical process control Log anomaly detection Embedding-based clustering Root cause analysis Correlation across metrics/logs/traces Auto-remediation Runbook → agent Capacity planning Predicted growth → suggest right-sizing
Koneksi ke Vault:
11. Decision Framework
11.1 Pilih Layer Berdasarkan Constraint
Budget limited + Predictable load → Bare-metal
Utilization matters + Some burst → Virtualization
Variable load + CapEx budget zero → IaaS
Microservices + K8s needed → Container
Event-driven + Burst to zero → Serverless
Latency <50ms globally → Edge
Multi-region + Compliance → Multi-cloud
Complex orchestration + Self-heal → AIOps
11.2 Anti-Pattern
Anti-pattern Fall-out Serverless monolith Function dependency hell Container everything tanpa state mgmt Data loss risk Cloud-only, no exit plan Lock-in biaya tinggi Edge without origin Inconsistency antar region Bare-metal dalam startup phase Time-to-market lambat
12. Trade-off Matrix
Layer Setup Time Ops Complexity Cost (Idle) Cost (Peak) Latency 0 Bare-metal Weeks Highest $0 (asset) $0 (asset) Sub-ms 1 Virtualization Hours High Low Low Sub-ms 2 IaaS Minutes Medium $0 (deallocated) Medium <50ms 3 Container PaaS Minutes Medium-High $0 (scaled to 0) Medium-High <100ms 4 Serverless Seconds Zero $0 Pay-per-invoke 50-500ms 5 Edge Seconds Medium $0 Pay-per-invoke <50ms 6 Multi-cloud Hours High Variable Variable Variable 7 AI-Orchestrated Variable Low (managed) $0 Variable Variable
13. Cross-Reference ke Vault
References
T. Erl et al. “Cloud Computing: Concepts, Technology & Architecture.” 2013.
B. Burns. “Designing Distributed Systems.” O’Reilly, 2018.
Kubernetes Authors. “Kubernetes Documentation.” https://kubernetes.io/docs/
CNCF. “Cloud Native Trail Map.” https://github.com/cncf/trailmap
A. Wiggins. “The Twelve-Factor App.” 2012.
M. Kavis. “Architecting the Cloud.” Wiley, 2014.
Brendan Burns et al. “Kubernetes: Up and Running.” O’Reilly, 2022.
Kelsey Hightower et al. “Kubernetes the Hard Way.” 2020.
Liz Rice. “Container Security.” O’Reilly, 2020.
Linux Foundation. “Hyperledger Architecture.” 2020.
NIST. “NIST SP 500-325: Fog Computing Conceptual Model.” (2018).
Adrian Cockcroft et al. “Migrating to Cloud-Native Application Architectures.” 2015.
A. W. Services. “Well-Architected Framework.” (2024).
Gartner. “Magic Quadrant for Cloud Infrastructure.” (2025).
IDC. “Future of Digital Infrastructure.” (2025).