AEM and Akamai: CDN Configuration and Image Optimisation

AEM and Akamai: Configuring a CDN with Image Optimization for AEM requires more than placing a publish instance behind a fast edge network. The design must connect AEM’s authoring and publishing workflow with Akamai’s caching rules, image delivery strategy, security controls, and invalidation process. When these parts agree, Australian visitors receive responsive media quickly while AEM teams retain control over content releases.

This pattern suits organisations serving customers across Sydney, Melbourne, Brisbane, Perth and regional areas from a smaller number of origin environments. A visitor on an NBN connection in Adelaide may have a very different route to the origin from a mobile user in Darwin, so edge caching and efficient image variants can have a visible effect on conversion, accessibility and hosting cost.

Map The AEM And Akamai Delivery Path

A typical implementation places Akamai in front of AEM Dispatcher, which then protects the publish tier. The request usually travels through DNS to an Akamai property, where the edge checks its cache before forwarding a miss to Dispatcher and the AEM publish farm. The origin should be reachable only through approved Akamai addresses or a private connectivity model, with TLS configured from the browser to the edge and from the edge to the origin.

Separate hostnames make the model easier to reason about. A public content domain can serve pages and client libraries, while an asset or media domain handles images, documents and other large files. AEM author should remain private, and replication agents should publish content to the intended publish tier rather than exposing authoring endpoints through the CDN.

A useful extension point is asynchronous processing. For example, an AEM workflow can place a transformation request on a queue and use AEM Lambda processing for tasks that should not delay an editorial activation. The result can be stored as a renditions package or delivered through a controlled asset pipeline before Akamai begins caching it.

Configure Origin, Hostnames And Cache Behaviour

In Akamai Property Manager, define the origin hostname, origin port, certificate requirements and forwarding rules before tuning cache performance. Preserve the host header when AEM uses it to select a virtual host, and forward only the headers and cookies required by the application. Passing every browser cookie to the origin can create needless cache fragmentation and expose session information to paths that should be public.

Cache rules should distinguish HTML, client libraries, images, JSON endpoints and administrative paths. Public images can generally receive a long time-to-live, while pages containing frequently changing stock, pricing or campaign information need a shorter policy. Use Cache-Control and, where supported by the Akamai configuration, surrogate controls so the edge has explicit instructions rather than relying on defaults.

Dispatcher remains an important second cache. Its filters should block selectors, unwanted extensions, query-string abuse and internal paths. A common mistake is to tune Akamai while leaving Dispatcher unable to cache the same URL patterns. Test a cache miss at both layers, then test a hit at Akamai, a hit at Dispatcher and a request that must reach AEM.

For teams reviewing event material and implementation references, the APAA conference archive can sit alongside technical session recordings as a reminder that architecture decisions benefit from both platform detail and practitioner experience.

Build A Practical Image Optimisation Pipeline

Image delivery begins in AEM Assets. Establish conventions for filenames, metadata, focal points, alt text and permitted formats, then create renditions for the actual display contexts used by the site. A hero image for a wide desktop layout should not be sent to a narrow mobile card at its original dimensions. Width-based renditions, responsive srcset markup and a suitable sizes attribute reduce transfer without damaging visual quality.

The edge can complement AEM’s rendition system, but it should not become an uncontrolled image manipulation layer. If Akamai Image Manager or another image transformation capability is used, define an allow-list of operations such as width, quality and format. Reject arbitrary remote origins and unbounded dimensions, since attackers can otherwise generate many expensive variants and fill the cache with low-value objects.

Modern format negotiation needs careful testing. WebP and AVIF can reduce payloads, but the application must send the correct Content-Type, vary responses when negotiation depends on the Accept header, and provide a reliable fallback. A consistent URL strategy is often easier to cache than a large collection of query-string combinations. Where the same image is requested in many sizes, normalise parameters and remove irrelevant tracking values before the cache key is built.

Image optimisation should include editorial governance. Compressing every file aggressively may harm product photography, artwork or maps, while leaving every upload untouched increases page weight. AEM workflows can inspect dimensions and file size, generate approved renditions, and flag assets that need human review.

Control Invalidation, Security And Personalisation

Publishing an AEM page does not automatically mean every related object has disappeared from Akamai. A page may reference a newly activated image, client library or content fragment while the old response remains at the edge. Define a purge sequence that covers the page, dependent assets and any API response affected by the activation. Fast purge is valuable, but predictable versioned URLs are often safer for immutable files such as hashed JavaScript and CSS.

Use cache tags, surrogate keys or a clear URL inventory where available. A release process should record what was activated, which paths were purged, and whether the purge completed successfully. Emergency invalidation must be protected by strong authentication and role separation; a broad purge endpoint exposed to the public internet is a serious operational risk.

Personalised responses require a hard boundary. Do not cache account pages, cart data or responses containing sensitive customer information as if they were public marketing pages. For experimentation, cache the stable shell and resolve the variation through a controlled client or edge mechanism. AEM teams considering AEM and Adobe Target should document whether the test decision occurs at the edge, in the browser or during origin rendering.

Australian sites should also consider privacy obligations under the Privacy Act and the Australian Privacy Principles when cookies, identifiers or behavioural segments cross the edge. Retail campaigns around Boxing Day or end-of-financial-year sales can generate sharp traffic spikes, making an explicit separation between public campaign content and logged-in experiences especially important.

Connect AEM Workflows With Content Intelligence

AEM can manage the editorial source, metadata and publishing state, while specialist services handle analysis or recommendation logic. Image metadata may support search and accessibility, and content signals can help select the right asset for a campaign. Any enrichment process should return a traceable result to AEM rather than silently changing an asset in production.

Machine learning may identify subjects, generate tags or rank content, but the output needs confidence thresholds and editorial review. A useful starting point is the machine learning personalisation discussion, then adapting the concept to the organisation’s consent model, data quality and Australian customer base.

For rule-heavy classification, a declarative approach can be easier to audit than a collection of hidden scripts. Teams experimenting with policy or relationship analysis may find Soufflé Datalog relevant for expressing facts and rules separately from the application that executes them. It is not a replacement for AEM workflows, but it illustrates how explainable processing can support asset governance.

Keep asynchronous services outside the synchronous page request wherever possible. A queue, retry policy, dead-letter path and idempotent job identifier prevent a temporary analysis failure from blocking publication. Editors should see whether processing is pending, complete or failed, with enough detail to correct an asset without involving an engineer.

Test Performance And Operate The Edge

Performance testing should represent real Australian conditions rather than a single office connection. Run journeys from Sydney, Melbourne, Perth and at least one regional location, comparing cold-cache and warm-cache behaviour. Include mobile devices, slower links and pages with many editorial images. Measure time to first byte, largest contentful paint, image transfer size, cache hit ratio and origin request volume.

Use Akamai logs, AEM request logs, Dispatcher statistics, browser performance data and synthetic monitoring together. A high edge hit ratio can still hide poor user experience if the wrong image rendition is cached or if HTML waits on a slow third-party script. Watch for cache-key explosions caused by query parameters, cookies, inconsistent trailing slashes and alternate hostnames.

Release testing should cover activation, purge, replacement of an existing asset, rollback and deletion. Confirm that an unpublished image cannot be retrieved through an old cached URL, while immutable assets remain available for previously cached pages when appropriate. Test failed origin responses too: a controlled stale response can preserve availability, but it must never serve private or incorrect content.

Document ownership across the AEM, Akamai, networking and editorial teams. Include certificate renewal, origin allow-list updates, purge permissions, image policy changes and escalation contacts. Clear runbooks matter during a sale launch in Melbourne or a service disruption affecting Perth users, when quick diagnosis is more valuable than an elegant but undocumented configuration.

Apply this design to a representative AEM site: place the publish tier behind Akamai, define separate cache policies, generate responsive renditions, validate purge behaviour and measure results from Australian locations. Review the recordings and linked technical material as your team builds the implementation, then turn the tested rules into version-controlled configuration and an operational runbook.