BLUEPRINT // DOKPLOYHUB // REV 2026.11 Self-Hosted PaaS Architecture VI · Tiếng Việt
DH DokployHub

DOKPLOYHUB · METHODOLOGY · CONTAINER MEMORY & TCO EQUATIONS

DokployHub Engineering Methodology: Container RSS, CI Build Spikes & SaaS TCO Physics

Every hardware recommendation and SaaS bill comparison on DokployHub is computed deterministically from documented Linux cgroup v2 Resident Set Size (RSS) baselines, V8/Cargo compilation heap formulas, and published vendor rate cards retrieved 2026-11-12.

We publish every equation used by DokployHub openly so systems engineers can audit our numbers against /sys/fs/cgroup/memory.current, BuildKit heap profiles, and 2026-11-12 public vendor rate cards.

DokployHub Deterministic Memory, Disk & TCO Equations

  1. 1. Control-Plane + Runtime RSS Memory Equation (DokployHub)

    Total baseline host RAM equals Linux kernel + systemd/Docker daemon overhead (0.45 GB) plus the PaaS orchestrator control plane (0.65 GB for Dokploy + Traefik + Postgres + Redis, 1.35 GB for Coolify v4 + Soketi + Dragonfly, 0.40 GB for Kamal 2 proxy, or 1.15 GB for Dokploy Swarm Manager) plus (N × per-app steady RSS) plus shared PostgreSQL/Redis buffer pools.

  2. 2. Concurrent CI Build Heap Spikes & Exit Code 137 OOM Prevention (DokployHub)

    In-place Git deployments do not fail at idle—they fail when Nixpacks, Railpack, Webpack/Turbopack, or rustc allocate 2.2 GB to 3.2 GB of transient anonymous memory on top of running apps. Our sizer adds the stack-specific CI build spike multiplied by the build topology factor (1.0x for in-place VPS builds, 0.18x for SSH Remote Build Servers, and 0.12x for GHCR pre-built image pulls) plus an 18% Linux filesystem page-cache headroom buffer before snapping to KVM hardware tiers (4, 8, 16, 32, or 64 GB).

  3. 3. Docker Overlay2 & BuildKit NVMe Disk Sizing (DokployHub)

    Self-hosted PaaS hosts retain active image layers, rollback replicas, BuildKit build caches, container JSON logs, and PostgreSQL WAL segments. Recommended NVMe capacity reserves 24 GB for Ubuntu 24.04 + base images, 18 GB for in-place BuildKit layer cache, and 2.8 GB to 5.5 GB per deployed application service.

  4. 4. SaaS Replacement TCO Baseline (Vercel Pro, Railway, Render & Heroku) (DokployHub)

    Monthly SaaS comparisons average three real-world cloud baselines for the selected container count: (1) Vercel Pro ($20/seat for 2–5 team seats + Managed Postgres/KV + Fast Origin Transfer egress overages), (2) Railway Pro / Render Standard (seat fees + metered GB-month RAM, vCPU, and $0.05–$0.10/GB egress), and (3) Heroku Standard-1X dynos ($25/service) + Standard-0 Postgres ($50/mo) + Mini Redis ($15/mo), contrasted against flat-rate KVM NVMe compute + $6/mo 1TB S3 off-site backup.

  5. 5. Collective Governance & Commercial Firewall (DokployHub)

    All DokployHub sizing equations are maintained online by the DokployHub PaaS Repatriation Collective as a digital-only engineering team rather than a personal blog, ensuring vendor-neutral reproducibility. Outbound /go/ links keep DokployHub free with zero login walls while partner commission rates are never an input variable in any RAM equation, OOM threshold, or TCO sort order.

dokployhub-ram-equation.txt 
Peak_Deploy_RAM_GB = (0.45_Kernel + ControlPlane_RSS + N_Apps * PerApp_RSS + Shared_DB_Pool + Build_Spike * CI_Topology_Factor) * 1.18
Recommended_NVMe_GB = max(50, ceil(24_OS + N_Apps * PerApp_Disk_GB + BuildKit_Cache_GB))
DokployHub Public Vendor Rate Citations (Retrieved 2026-11-12): Rates reflect publicly listed plans on Hostinger VPS Hosting, Vultr Cloud Compute Pricing, UltaHost VDS Hosting, InterServer VPSSlices, and ChemiCloud NVMe Hosting. Notice: vendor pricing, promotional credits, and renewal rates are subject to change at checkout without prior notice.

DokployHub Pre-Cutover Verification Principles (Engineering Methodology)

Across our deterministic DokployHub runbooks, the DokployHub PaaS Repatriation Collective advises rehearsing a full pg_dump or Litestream SQLite WAL restore from off-site S3 storage onto a staging container before lowering DNS TTLs to 60 seconds for Engineering Methodology.

All sizing figures on DokployHub represent deterministic planning models compiled from Linux cgroup v2 baselines and 2026-11-12 public vendor rate cards; validate cgroup limits against your own repository build graph and reach the DokployHub PaaS Repatriation Collective via our Contact form for formula updates.