Back to blog

Alibaba Cloud CDN vs ESA: Choosing the Right Option for Static Sites and Dynamic Workloads

CdnChart Technical TeamPublished on 2026-11-2117 min read4,473
Alibaba Cloud CDN vs ESA: Choosing the Right Option for Static Sites and Dynamic Workloads

How should you choose between Alibaba Cloud CDN and ESA? For a newly built small static website, it is advisable to make the ESA Basic plan an early comparison option: it also supports static content delivery, and the Basic plan does not charge separately for HTTPS requests. For websites already using CDN effectively, the first step should be to review actual bills and migration costs. For services that must be served from the origin, such as login, orders, and real-time queries, it is more appropriate to evaluate ESA's dynamic route optimization and security capabilities; however, if the slowness comes from the database or application queuing, switching acceleration products will not solve the main problem.

The decision order should be: whether content can be shared, where the time is spent, which specific functions are needed, and only then the plan price. A static website does not automatically rule out ESA, and a dynamic website does not mean you must buy the Standard plan. Only by separating these two points can you avoid buying a more expensive plan while the original problem remains.

How to choose between Alibaba Cloud CDN and ESA: start by breaking it down request by request

“Corporate website,” “e-commerce website,” and “static site” are only overall labels; they cannot directly determine how each request should be handled. Even if a front-end project is deployed entirely as static files, it may still call login, search, and payment APIs after it opens. A server-generated article page may also be suitable for shared caching if it returns the same content to all visitors and allows a short delay before updates.

Start by listing actual requests from the browser network panel or access logs. At a minimum, record the hostname, path, whether identity is carried, content update requirements, request count, and bytes transferred. Do not count only the homepage, and do not calculate a single hit rate for both admin APIs and images. The three types of requests below have completely different purchasing rationales.

Request type

Problem to solve

Selection priority

Images, versioned scripts and styles, public download files

Reuse the same content to reduce origin fetches and download wait times

Compare caching configuration, delivery performance, and actual total cost of CDN and ESA

Public article HTML, public product information, or configuration APIs

Share responses within the allowed update delay

First confirm cache conditions and invalidation methods; do not judge based only on file extension

Login, user orders, balances, personalized prices, and write APIs

Keep responses isolated and optimize each path that must reach the origin

Evaluate ESA network and security features while checking application latency

Request counts, byte counts, and business importance should also be considered separately. A large number of small images may account for most requests but not most bandwidth; an inventory API may carry very little traffic but determine whether a user can place an order. The former affects request billing and cache efficiency, while the latter affects transaction completion. Using “static requests account for 90%” to directly conclude “only static acceleration is needed” can easily overlook the one request that is actually slowing down the business.

This also means the value of the two products may lie in different places. Static delivery mainly seeks reuse benefits, dynamic optimization mainly seeks link quality, and security capabilities reduce the interference of malicious requests with normal business. Attributing all three types of benefits to “average page load speed” can easily blur the selection criteria.

Static websites can also choose ESA; the focus is cache reuse and update methods

Alibaba CloudESA plan feature comparisonlists that the Basic plan supports global static content delivery, edge cache duration, query string configuration, purge and instant prefetch, among other capabilities. Therefore, for a new static site, there is no need to exclude ESA from the candidates just because its name contains the word “security.”

For static resources, the first thing that affects the benefit is whether later requests can use copies left by earlier requests. If the same image is constantly requested with random parameters, or each user uses a completely different download address, the same file may be split into many cache objects. If this access pattern continues after switching products, requests will still repeatedly go to the origin. You should first distinguish which parameters change the content and which are only used for analytics, then organize cache keys without breaking authentication and version isolation.

The second issue is the update mechanism. Scripts and styles are suitable for content version numbers: when the content changes, generate a new address. Entry HTML requires more careful expiration and purge scheduling. Setting a very long cache duration for all files may cause HTML to reference old resources that have been deleted, or cause users to keep seeing an old version. Buying a higher-tier plan cannot automatically fix the release order.

A practical approach is: first upload the new version of resources, confirm that they are accessible from both the origin and the acceleration path, then update the entry page, and keep a compatibility window for resources that the old entry page may reference. When public HTML needs to be cached, specify which content may be briefly stale and how it should be invalidated after updates. For pages whose membership status, regional pricing, or post-login menu differs, do not simply apply the rules for purely static articles.

If an existing CDN has stable cache hits in the target region, reasonable origin load, and a mature release process, keeping it is a reasonable choice. CDNChart's public URL probes can help check access paths and regional differences, but they cannot replace checks of content versions, cache status, and full page loading. When selecting a solution for a static website, first ensure correct delivery, then compare cost and experience benefits.

Dynamic acceleration improves network wait time; application processing must still be assessed separately

The fact that dynamic requests cannot use shared caching does not mean there is no value in passing through edge nodes. ESA'sintelligent routinguses edge network probing and dynamic route selection to optimize transmission paths, making it suitable for evaluating link issues across regions and carriers. It changes how requests reach the application; it cannot perform database queries on behalf of the application, nor does it automatically eliminate connection pool queuing.

A latency budget can be used to clarify the upper limit. Suppose an API's time to first byte is 600 milliseconds. After separating non-overlapping stages, application processing accounts for 450 milliseconds, and the remaining network, connection, and proxy wait totals 150 milliseconds. Even if the latter half is reduced by half overall, total wait only drops to 525 milliseconds, a 12.5% improvement. If application processing is reduced from 450 milliseconds to 200 milliseconds while the other parts remain unchanged, total wait becomes 350 milliseconds, an improvement of about 41.7%. This set of assumptions is used to explain the optimization headroom and does not represent a performance commitment from ESA.

Therefore, if access near the origin is also slow and application logs show the same query consistently taking time, optimize the query, indexes, connection pool, or queuing mechanism first. Conversely, if origin processing is stable but distant users slow down noticeably on specific carriers at specific times, dynamic path optimization is closer to the actual problem. If both are slow, split application and network into two tasks and address them in parallel.

Also note that “origin fetch” does not mean “establishing a new connection every time.” Whether connections can be reused, how the origin protocol is configured, and whether the origin can handle concurrent connections all affect the additional overhead of dynamic requests. For an already short and stable same-region path, adding a proxy hop may not be worthwhile; for a long-distance path with obvious fluctuations, reducing failures and tail latency may be more valuable than shaving a few milliseconds off the average.

Dynamic acceptance testing must check real operations. If an API returns 200 but returns a challenge page, login page, or incomplete data, the business still fails. Public GET requests, authenticated queries, write requests, and long-lived connections should be checked separately; you cannot use the download result of a static image to make decisions for the entire application. For the difference between the two mechanisms, you can usethe division of responsibilities between API caching and dynamic accelerationto determine the API strategy.

The Basic plan already supports intelligent routing; the Standard plan should be paid for specific capabilities

ESA Basic supports intelligent routing, but intelligent routing is billed separately. You should not interpret “the Basic plan can enable it” as free, nor should you interpret the Standard plan's API acceleration positioning as meaning the Basic plan cannot handle dynamic business at all. When choosing a plan, separate three items: whether a feature is supported, whether the rule quota is sufficient, and how it is charged once enabled.

According to the public feature table as of November 21, 2026, the following differences between the Basic and Standard plans are more likely to affect the actual use of small and medium-sized websites. They are more suitable as reasons for upgrading than the vague claim of “stronger capabilities.”

Required capability

Basic plan

Actual change in Standard plan

Cache rules

10 rules, sharing the corresponding configuration quota

50 rules, suitable for sites that need more business exceptions

Cache analytics, scheduled prefetch

Neither is supported; instant prefetch is supported

Both capabilities are supported

WAF managed rules and scan protection

Managed rules support only basic rules; scan protection rules are not supported

Supports all managed rules and the corresponding scan protection configuration

Instant logs

Not supported; offline logs are available

Supported, making it easier to troubleshoot issues around the current request

Intelligent routing

Supported; billed separately once enabled

Also supported; still subject to the corresponding value-added fee

For example, for a small site with only images and public articles, ten rules are enough to express the main business logic, and offline logs can meet routine checks. There is little basis for upgrading to Standard solely for “faster” performance. For another site that is frequently scanned and needs complete managed rules as well as instant investigation of false blocks, the value of the Standard plan is much more specific: it reduces the need to write custom rules and the wait for troubleshooting, rather than simply providing more traffic.

Do not treat a paid tier as a pass for all security capabilities. Bot management, full API security, and enterprise-grade protection each have their own functional boundaries; a plan supporting API management does not mean it has all API security capabilities. Buy according to the response actions and quotas you actually need, and do not infer from the words “has WAF” or “supports Bot” that all attacks can be handled.

The Free plan can be used to test non-core business, but the official positioning is for testing purposes, and it does not provide a committed SLA or after-sales ticket support. If core business requires these guarantees, budget for a paid plan. This judgment is based on the scope of service, not on the idea that “free must be slow” or “paid must be fast.”

Billing the same set of bytes: CDN and ESA use different measurement bases

On November 21, 2026, Alibaba Cloud'sCDN detailed pricing pagelists the pay-as-you-go traffic price for mainland China. The first tier, 0 to 10 TB, is CNY 0.24/GB, and traffic is converted in base 1024; other regions and higher tiers have separate prices. In addition,static HTTPS request billingspecifies that the first 5 million requests each month are free, and the excess is CNY 0.05 per 10,000 requests. Downstream traffic packages cannot offset this request fee.

ESA'sBasic plan feeis subscription plus overage: the Basic plan is CNY 9.9/month and includes 50 GB, with overage at CNY 0.198/GB; the Standard plan is CNY 375/month and includes 500 GB, with overage at CNY 0.75/GB. The Basic plan does not charge separately for ordinary static, dynamic, or HTTPS requests; value-added items such as intelligent routing are billed separately. ESA converts traffic in base 1000, and measurement includes request and response traffic between the client and the edge.

The following assumes one site's business volume for a full calendar month: response downstream traffic under the billing basis is 400,000,000,000 bytes, request upstream traffic is 10,000,000,000 bytes, and static HTTPS requests are 20 million; all CDN traffic occurs in mainland China. Assume the account has no other usage consuming the HTTPS free quota, does not purchase resource packages, does not use promotions or contractual discounts, and does not enable intelligent routing or other value-added items. Origin and storage fees are calculated separately.

Thus, the traffic used by CDN to calculate downstream fees is 400,000,000,000 divided by 1024 cubed, approximately 372.5290 GB; ESA's combined request and response traffic is 410,000,000,000 bytes, or 410 GB. Here, the network bytes already aggregated under the billing basis are used directly, without adding protocol overhead again. The two products cannot simply apply the same “400 GB” figure to pricing.

Plan

Cost calculation in this example

Monthly cost

CDN mainland China pay-as-you-go

Downstream fee approx. 372.5290 × 0.24; HTTPS overage of 15 million requests, plus 1500 × 0.05

Approx. CNY 164.41

ESA Basic

9.9 + (410 − 50) × 0.198

CNY 81.18

ESA Standard

410 GB is within the 500 GB allowance for a full month

CNY 375

Keep precision for intermediate usage, and round the cost to two decimal places at the end. This example illustrates two points: for request-intensive static business, the CDN HTTPS request fee cannot be omitted; and the Standard plan is not intended to obtain cheaper traffic. After usage exceeds the plan allowance, the publicly listed overage unit price of the Standard plan is higher than that of the Basic plan. The reason to upgrade should be the features and service capabilities you actually need.

If intelligent routing is enabled, it is billed separately at CNY 0.1 per 10,000 requests. For example, if 100,000 requests are actually counted for this service every day for 30 consecutive days, the cost is CNY 1 per day, or CNY 30 in total. Here, intelligent routing billable usage is used; the total number of website visits is not directly treated as its billable count. This item is settled daily, and you should reconcile it against the actual effective rules and bill.

The included traffic for the first and last months of a plan is also prorated according to the effective duration, and unused allowance in the current month does not carry over. Existing CDN resource packages cannot be directly used to offset ESA charges, and related origin or storage-side costs will not automatically disappear just because you switch to ESA. The above comparison is suitable for observing the pay-as-you-go model; users with existing discount packages or contractual prices should substitute their actual unit prices rather than treating the list-price difference as a guaranteed saving.

For login, orders, and user APIs, get cache isolation right first

One of the biggest changes when moving from CDN to a full-site proxy is that requests that previously did not pass through the acceleration layer may also fall within the scope of rules. ESAdefault cache rulesdetermine cache eligibility based on request method, matching rules, and resource type, and by default mainly target the corresponding file extensions. Custom rules can change how requests are handled, so “by default, certain pages are not cached” cannot replace the security boundary after launch.

For login results, personal profiles, orders, and personalized quotes, it is advisable to first use explicit rules to bypass shared caching, while the origin also returns appropriate cache controls. Do not ignore all Cookies, delete all query parameters, or force a long cache duration for the entire domain just to improve the hit rate. If authentication parameters and identity information participate in content selection, dropping them may mix responses from different users.

Nor should you assume that a request is necessarily protected just because it carries a Cookie, or that a response is necessarily suitable for caching just because it returns JSON. A public product list can be JSON, and a personal balance can also be JSON; the real distinction is whether the same request should provide the same content to different users. For the specific meaning of cache control directives, seeShared, private, and no-store semantics of Cache-Controland verify them together with the actual priority of edge rules.

During acceptance testing, prepare two test accounts with different permissions and one anonymous session, and cross-request the same paths. Check whether user fields, permissions, and data correspond; whether logged-in content can still be retrieved after logout; and whether invalid credentials are correctly rejected. If user mix-ups, unauthorized data, or responses that should not be reused occur, pause traffic expansion and fix the rules first. The loss from such errors is far greater than a few missed cache hits for an image.

Existing CDN does not need to be dismantled all at once; you can divide responsibilities by hostname with ESA

For sites that are already running stably, a more appropriate starting point is often to keep static domain names and let only the business that truly needs dynamic optimization enter ESA separately. For example, static.example.com continues to deliver images, scripts, and styles, while api.example.com handles business APIs that are not cached. Connect the two hostnames to the appropriate services respectively; responsibilities are clear, and it is easier to compare failures and costs.

ESA supports both CNAME and NS onboarding. If you want only some subrecords to use ESA while other records continue to be handled by the original service, consider CNAME onboarding first and retain existing DNS management; if you want unified management of site DNS and acceleration, then evaluate NS onboarding and record migration. Do not move unrelated email, verification records, and other business domains together just to test the benefit of one API.

Dividing responsibilities by hostname does not mean filling in two CNAMEs for the same domain at the same time. It is even less advisable, without a clear architectural purpose, to point ESA's origin directly to another CDN and then point that CDN back. Chaining proxies increases the difficulty of troubleshooting rules, Host, certificates, and log ownership, and incorrect settings may create loops. By default, prefer parallel division of responsibilities rather than stacking layers.

After domains are separated, also check the allowed origins for APIs, Cookie scope, login redirects, and certificate coverage. A front end that previously worked on the same domain may need adjustments to cross-origin and session policies after it calls a separate API domain. These are architectural costs and should be listed before purchase; do not wait until after the switch to discover that every call reports an error.

For the original CDN path, you can usedomain and cache effectiveness check stepsto keep a baseline. During the pilot, retain DNS, certificate, and origin configurations that can be restored, and control the scope of changes. In this way, even if dynamic acceleration delivers insufficient benefits, you can end the pilot without having to rebuild the entire static release pipeline along the way.

Accept static and dynamic groups separately; do not decide solely by an average homepage score

For the static group, select images, scripts, and entry pages that actual users will access, and fix the URL, content version, region, carrier, and time period. Observe the first origin fetch and subsequent cache hits separately, and retain status codes, actual connected IPs, cache evidence, and content verification. Changing random parameters to force “new requests” may turn the test object into a cold cache and is not suitable for proving everyday cache-hit experience.

For the dynamic group, select real business paths covering at least public queries, authenticated queries, and necessary write operations. You must confirm that both paths execute the same business: status codes, response fields, permission results, and application processing stages should be comparable. For operations such as payments and order placement, use a business-approved test environment or controlled test accounts; do not repeatedly submit real transactions just to collect samples.

For initial regional screening, you can useCDNChart website speed testto view the response IP, status, total time, and the DNS resolution, connection, and download stages displayed on the page for a public URL. It is suitable for discovering access failures or abnormal connection waits in certain regions, but it cannot replace login session verification, full page rendering, stress testing, and application logs. API experience also requires browser network records and origin traces to be interpreted together.

For groups with sufficient request volume, look at the median, slower requests, and failure rate separately, and list peak evening hours separately. Do not mix different regions and cache states into one average, and do not count only successful requests. Even if a new solution makes successful requests faster, if failures or business false blocks increase, traffic should not be expanded. A small number of probes can only provide clues; they cannot be used to claim that a product is faster in all user regions.

  1. If static cache hits are still poor: first check cache keys, expiration, and update methods; do not treat a plan upgrade as a fix for now.

  2. If origin processing is stable but a specific remote link shows obvious waiting and failures: evaluate ESA intelligent routing for this group of business and observe whether the same group improves.

  3. If the origin application stage accounts for the majority: optimize the application first, keep the original acceleration path as a baseline, and validate network capabilities separately.

  4. If new security policies falsely block normal requests: identify the specific rules and narrow the scope of adjustments, verify that business has recovered, and do not replace troubleshooting with turning off protection entirely.

For the three types of business, spend the budget where it is closest to the problem

Newly built documentation sites, image sites, or personal websites:If they mainly serve shareable static resources and have simple rule and log requirements, first compare ESA Basic with the actual CDN quote. Under the public pay-as-you-go scenario above, the Basic plan has a clear cost advantage and is worth piloting first; acceptance should focus on target regions, cache reuse, and update correctness, rather than buying Standard first. If you already have low-cost resource packages, mature automation, or other required features, keep CDN and calculate the actual benefit of switching.

E-commerce and membership websites that already have static delivery:Keep the static CDN that performs well, and evaluate dynamic entry points such as login, queries, and orders separately. If network paths are unstable, validate intelligent routing first; pay for Standard only when you truly need more rules, full WAF managed rules, scan protection, or instant logs. Cache correctness and normal user reachability are prerequisites; an increase in hit rate cannot prove that order APIs are suitable for shared caching.

Backend or SaaS systems with slow queries:If access near the origin is also slow and application records show that most time is spent on queries, lock waits, or internal dependencies, fix the application first. ESA can take on appropriate access and security responsibilities, but a plan upgrade should not be treated as a way to shorten database execution time. Only after the application stage is stable and the network portion still continuously affects target users does it make sense to increase the routing optimization budget with a clear acceptance target.

The difference among these three approaches lies in which highest-cost problem is addressed first. Static sites value reusable bytes and request costs, transaction sites value identity isolation, failures, and security response, and backend systems value application execution. Purchasing the same brand does not mean every type of business should take the same path and enable the same set of features.

Before deciding to migrate, calculate feature benefits and transformation costs again

The difference in the price list is not yet the migration benefit. Existing CDN users need to include the work hours for domain and certificate adjustments, rule migration, cache rebuilding, release process changes, log adaptation, and rollback drills. If you save only a small monthly fee but have to disrupt a stable release pipeline, adopting ESA first on new projects or independent subdomains is usually more appropriate than replacing the entire site.

Feature benefits should correspond to observable outcomes. Buying cache analytics is to find non-reusable objects faster; buying more complete security rules is to reduce malicious requests and control false blocks; buying instant logs is to shorten the time to locate the current fault. Write these goals down clearly so you know whether the cost of a higher-tier plan is paying off, rather than buying it and then only looking at one total traffic curve.

For new static business with simple requirements, compare ESA Basic first; for a static CDN that runs well and has a suitable actual price, keep it and continue optimizing; for dynamic business that must reach the origin, first evaluate whether ESA improves the target link, then upgrade according to the exact features needed. If origin application processing accounts for most of the time, fix the application first. Only when cost, delivery correctness, and business outcomes all support a change is it worth expanding the scope of migration.

  • Alibaba Cloud CDN vs ESA
  • ESA Basic and Standard editions
  • Alibaba Cloud static website acceleration
  • ESA dynamic acceleration
  • ESA smart routing pricing
  • CDN vs ESA billing comparison

Related posts