14 · Dynamic content acceleration and Argo Smart Routing

Distinguish origin processing from network-path latency and choose the right layer among cache, Tiered Cache, and paid Argo.

CDN · PHASE 5ADVANCEDAbout 28 minutesDynamic · Argo · Origin Analytics

Choose the correct optimization layer first

Cache public repeatable responses. Use Tiered Cache to reduce the number of regional misses that reach the origin. Evaluate Argo when a request must reach the origin and the long-distance network path is unstable. If the database, an external API, or application computation dominates, optimize the origin first.

Where a dynamic request spends time

flowchart LR
  V[Visitor] --> E[Cloudflare edge]
  E --> C{Safely cacheable?}
  C -- Yes --> H[Edge / Tiered Cache]
  C -- No --> N[Cloudflare-to-origin network]
  H -- MISS --> N
  N --> O[Origin application and database]
  O --> E

Argo Smart Routing optimizes path selection between Cloudflare's network and the origin. It cannot reduce database queries, server rendering, third-party model/API calls, or browser JavaScript execution. The same boundary applies to AI-assisted SaaS, CMS, image, and PDF tools.

Decide from evidence

Observation	First action	Argo fit
Many static-object misses	Fix Cache Rules, TTL, Tiered Cache	Low; reduce origin traffic first
Dynamic API spends two seconds in origin	Inspect database, code, and dependencies	Low; routing is not dominant
Distant regions have slow, volatile origin trips	Compare Origin Analytics by region and time	High; run an experiment
Origin is near most visitors	Measure benefit and cost first	Potentially limited
Global users, one-region origin	Measure Tiered and Argo separately	Often useful, for different goals

Origin Analytics starts when Cloudflare decides to contact the origin and stops on response headers. It includes DNS, TCP/TLS, request transmission, origin processing, and the network return; Argo/Tiered time is included when enabled. It is therefore usually higher than APM that measures only server processing. Their aligned difference can reveal network and platform-path time.

Build a comparable experiment

Baseline by route group

Separate public cached resources, sign-in APIs, database reads, uploads, transform-job status, and third-party APIs. Record p50/p95/p99 origin response time, origin 5xx, request volume, and egress. Do not average every route into one number.

Fix obvious application bottlenecks first

Align origin APM with Origin Analytics using the same UTC time, Host, Path, and Ray ID. If server processing consumes most of the interval, improve queries, connection pools, cold starts, or external APIs first.

Enable Argo and billing notifications

Argo is a paid, usage-metered add-on and requires a Billing Profile. Enable it under Traffic → Argo Smart Routing, then configure a Usage Based Billing notification with a defined observation window and budget ceiling.

Keep Tiered Cache as an independent decision

Tiered Cache reduces how many requests reach the origin. Argo selects a better network path for requests that still need it. They can work together, but log each enablement time and metric independently to attribute results.

Inspect Argo Analytics

Under Analytics → Performance, compare origin TTFB when a Smart Route was and was not used, plus geographic distribution. Detailed Origin Performance currently needs at least 500 Argo-routed origin requests in the previous 48 hours. Extend the observation window when data is insufficient instead of citing a marketing average.

Three realistic AI-built application scenarios

Scenario	Do first	What Argo can do
SaaS dashboard	Bypass shared cache; optimize database and API	Improve the network leg for distant users of a single-region origin
Public CMS plus admin	Cache public pages; bypass admin	Improve admin and cache-miss origin paths
Image/PDF processing tool	Cache static frontend; keep upload/job dynamic	Improve control requests, not the actual transform computation

Do not promise a fixed performance percentage

Benefit depends on visitor geography, origin location, congestion, dynamic-traffic share, and origin processing. Draw conclusions only from your zone's Analytics and real users, and include cost, errors, and reliability in acceptance.

Expand, observe, or stop

Expand when target-region p95/p99 origin response time improves repeatably, 5xx remains stable, and cost is below business value.
Continue observing when there are fewer than 500 Argo origin requests in 48 hours or daily variance exceeds the difference.
Stop when origin processing dominates, no target region improves repeatably, cost exceeds budget, or errors worsen.
Preserve Tiered Cache and cache rules on rollback; do not disable independent optimizations together.

AI workbenchHuman reviewed

Ask AI to separate network latency from application latency

I will provide Cloudflare Origin Analytics p50/p95/p99 by Host/Path/Region, routed and unrouted Argo TTFB, origin APM processing time, cache status, Tiered Cache state, request volume, 5xx, origin location, and Argo cost. Partition time into cacheable work, Cloudflare-to-origin network, origin application/database, and frontend. Decide whether the Argo experiment has enough samples and positive value. Do not cite a fixed speed percentage; define expand, continue-observing, and stop criteria.

Finish with end-to-end performance verification and troubleshooting to create a consistent evidence packet.