CDN Origin Server Optimization for Devoted Infrastructure

Optimizing your devoted server as a CDN origin is a special self-discipline than optimizing it for direct person visitors. The CDN handles concurrency and geographic distribution; the origin wants to reply reliably to CDN pull requests with right cache headers, compressed belongings, and minimal time-to-first-byte.

How CDN Origin Requests Differ from Direct Person Requests

When CDN is deployed in entrance of your devoted server, person visitors patterns change essentially:

Direct visitors (no CDN): Each person request hits your server. Excessive concurrency means many simultaneous connections. Response time straight impacts person expertise.

CDN-proxied visitors: CDN edge nodes cache responses and serve them to customers from geographic PoPs. Your origin sees CDN pull requests — largely cache misses for brand spanking new or expired content material. Concurrency is decrease however requests could also be much less predictable (cache expiry causes simultaneous pulls from a number of edge nodes).

Cache miss storms: When a high traffic asset expires concurrently throughout CDN nodes, a number of edge nodes request the identical useful resource directly. An origin server that’s sluggish to reply causes cache misses to queue up, doubtlessly serving stale content material or failing requests throughout the refill window. Cloudflare calls this “origin defend” — routing edge-to-origin visitors by means of a single intermediate node reduces the variety of simultaneous origin pulls throughout cache refresh.

Nginx Configuration for CDN Origin Efficiency

The Nginx configuration in your devoted server wants completely different tuning for origin service versus direct person service.

Employee processes and connections:

# /and so on/nginx/nginx.conf

# Match employee rely to CPU cores

worker_processes auto;

# Improve for origin serving high-traffic CDN pulls

occasions {

    worker_connections 4096;

    use epoll;

    multi_accept on;

}

# /and so on/nginx/nginx.conf

# Match employee rely to CPU cores

worker_processes auto;

# Improve for origin serving excessive-visitors CDN pulls

occasions {

    worker_connections 4096;

    use epoll;

    multi_accept on;

}

Keepalive for CDN connections: CDN edge nodes make repeated requests to your origin. Keepalive connections get rid of TCP handshake overhead on every request:

http {

    # Maintain connections to CDN alive

    keepalive_timeout 120s;

    keepalive_requests 1000;

    upstream origin_backend {

        server 127.0.0.1:8080;

        keepalive 64;  # Connection pool measurement

    }

}

http {

    # Maintain connections to CDN alive

    keepalive_timeout 120s;

    keepalive_requests 1000;

    upstream origin_backend {

        server 127.0.0.1:8080;

        keepalive 64;  # Connection pool measurement

    }

}

Gzip compression at origin: CDN nodes usually cache the compressed model and serve it straight. Configure Gzip on the origin for all compressible varieties:

gzip on;

gzip_vary on;

gzip_proxied any;

gzip_comp_level 6;

gzip_min_length 1024;

gzip_types

    textual content/plain

    textual content/css

    textual content/javascript

    software/javascript

    software/json

    software/xml

    picture/svg+xml

    font/woff2;

gzip on;

gzip_vary on;

gzip_proxied any;

gzip_comp_level 6;

gzip_min_length 1024;

gzip_types

    textual content/plain

    textual content/css

    textual content/javascript

    software/javascript

    software/json

    software/xml

    picture/svg+xml

    font/woff2;

Stage 6 balances compression ratio in opposition to CPU value. Ranges 7-9 present marginal extra compression at important CPU overhead — hardly ever price it for static belongings.

Cache-Management Headers: The Most Necessary Origin Configuration

CDN conduct is nearly solely decided by the Cache-Management headers your origin sends. Getting these proper determines whether or not your CDN caches aggressively (lowering origin load) or pulls steadily (including pointless origin requests).

Static belongings (CSS, JS, photos, fonts): Set lengthy max-age with fingerprinted URLs. When the file content material modifications, the filename modifications (webpack content material hash, and so on.), so caching aggressively is protected:

location ~* .(css|js|woff2|woff|ttf|jpg|jpeg|png|gif|ico|svg)$ {

    expires 1y;

    add_header Cache-Management "public, max-age=31536000, immutable";

    add_header Differ "Settle for-Encoding";

}

location ~* .(css|js|woff2|woff|ttf|jpg|jpeg|png|gif|ico|svg)$ {

    expires 1y;

    add_header Cache-Management "public, max-age=31536000, immutable";

    add_header Differ "Settle for-Encoding";

}

The immutable directive tells CDN and browsers by no means to revalidate this URL — the content material won’t ever change at this URL.

HTML pages: Brief or no cache. HTML modifications with content material updates and shouldn’t be aggressively cached until you’ve carried out correct cache invalidation:

location / {

    add_header Cache-Management "public, max-age=300, stale-while-revalidate=60";

}

location / {

    add_header Cache-Management "public, max-age=300, stale-while-revalidate=60";

}

stale-while-revalidate=60 permits CDN to serve a barely stale HTML response whereas fetching a contemporary copy within the background — reduces origin requests with out serving stale content material to customers for greater than 60 seconds.

API responses: Usually no-cache or brief TTL:

location /api/ {

    add_header Cache-Management "non-public, no-store";

}

location /api/ {

    add_header Cache-Management "non-public, no-store";

}

Origin Defend Configuration

An origin defend is a CDN characteristic that routes all edge-to-origin visitors by means of a single regional PoP somewhat than permitting all international edge nodes to drag straight out of your origin. The sensible profit: as an alternative of fifty edge nodes every requesting the identical asset when a cache expires, one defend node makes the request and distributes the cached response to the opposite edge nodes.

Cloudflare’s Tiered Cache implements this as a Sensible Tiered Cache topology that mechanically selects the most effective intermediate PoP based mostly in your visitors patterns. Allow it within the Cloudflare dashboard beneath Caching > Tiered Cache.

For origins on InMotion’s Los Angeles knowledge middle, Cloudflare’s West US tier gives the bottom latency between the defend and origin. For the Amsterdam knowledge middle, the Western Europe tier is suitable.

Serving Static Belongings from NVMe: I/O Configuration

Your devoted server’s NVMe drives serve static belongings to CDN edge nodes. NVMe I/O throughput is never the constraint for static file serving — community bandwidth usually hits limits earlier than NVMe does. However file system configuration impacts serving velocity:

Open file cache: Nginx caches file descriptors and stat() outcomes for frequently-served recordsdata:

open_file_cache max=10000 inactive=30s;

open_file_cache_valid 60s;

open_file_cache_min_uses 2;

open_file_cache_errors on;

open_file_cache max=10000 inactive=30s;

open_file_cache_valid 60s;

open_file_cache_min_uses 2;

open_file_cache_errors on;

This eliminates repeated open() and stat() system calls for a similar recordsdata — measurable enchancment for high-request-rate static serving.

Sendfile: Use the kernel’s sendfile() system name as an alternative of learn()+write() for file switch. Zero-copy knowledge motion:

sendfile on;

tcp_nopush on;  # Batch sendfile requires effectivity

tcp_nodelay on; # Disable Nagle after knowledge is distributed

sendfile on;

tcp_nopush on;  # Batch sendfile requires effectivity

tcp_nodelay on; # Disable Nagle after knowledge is distributed

AIO for big recordsdata: For video recordsdata, massive downloads, or different belongings over 1MB:

aio on;

directio 512;  # Use direct I/O for recordsdata over 512 bytes

output_buffers 2 128k;

aio on;

directio 512;  # Use direct I/O for recordsdata over 512 bytes

output_buffers 2 128k;

Monitoring Origin Efficiency from the CDN Perspective

Customary server monitoring reveals your server’s useful resource utilization. Origin efficiency from the CDN’s perspective requires monitoring from the CDN layer.

Cloudflare Analytics reveals origin response time (the time from CDN edge to your server and again), cache hit ratio, and error charges. Goal:

• Cache hit ratio: 85%+ for content-heavy websites, 95%+ for static asset heavy websites
• Origin response time (TTFB at edge): beneath 100ms for cached responses, beneath 500ms for origin pulls
• Origin error fee (5xx): beneath 0.1%

Nginx entry log evaluation reveals what the CDN is definitely requesting out of your origin. A excessive quantity of requests for a similar URL signifies a caching drawback — both the URL isn’t being cached or cache is expiring too steadily:

# Discover most steadily requested URLs (potential caching misses)

awk '{print $7}' /var/log/nginx/entry.log | type | uniq -c | type -rn | head -20

# Discover most steadily requested URLs (potential caching misses)

awk '{print $7}' /var/log/nginx/entry.log | type | uniq -c | type -rn | head -20

If the identical dynamic URLs seem repeatedly, they’re both not cacheable (right for API endpoints) or lacking Cache-Management headers (fixable).

WordPress as CDN Origin

WordPress websites generally run behind CDN with a devoted server as origin. WP Rocket, W3 Whole Cache, and LiteSpeed Cache all assist CDN integration and handle Cache-Management headers mechanically for WordPress-specific content material varieties.

Key WordPress configuration for CDN origin serving:

• Allow object caching with Redis (reduces dynamic web page technology time, enhancing origin TTFB)
• Set web page cache TTL to 30-60 minutes for logged-out customers
• Exclude admin URLs, cart, checkout, and account pages from caching
• Allow CDN integration within the caching plugin to rewrite asset URLs to CDN hostnames

InMotion Internet hosting’s Premier Care consists of InMotion Options consulting time for groups organising CDN integration for the primary time, that hour of month-to-month consulting time can be utilized for CDN configuration overview and optimization.

Associated studying: Community Latency Optimization for Devoted Servers | Server Useful resource Monitoring & Efficiency Tuning