TechPrufer
Get started
Cloudflare
Cloudflare logoCloudflare · Feature trail

Image Optimization Performance

4 posts · Jun 2012 – Sep 2023

What this trail covers

Cloudflare's image optimization capabilities have evolved to include support for mozjpeg 2.0, offering improved compression ratios and success rates compared to libjpeg-turbo. This involves a rewritten Go executable for Polish and ongoing optimization efforts for Intel processors. The platform also supports AVIF format and blur options for image variants.

All Cloudflare trails
  1. Image optimization made simpler and more predictable: we’re merging Cloudflare Images and Image ResizingThis post announces the merging of Cloudflare Images and Image Resizing into a single product. It introduces a new pricing model of $0.50 per 1,000 unique transformations for images not stored at Cloudflare, aiming for more predictable costs. A zero-storage plan is also introduced for users who only need image resizing. Existing Image Resizing customers can opt into the new pricing model.
  2. Cloudflare Images introduces AVIF, Blur and Bundle with StreamIntroduced AVIF support for Cloudflare Images by integrating AVIF format detection into the image delivery Worker, leveraging the Accept header to serve AVIF when supported by the client. Added a 'blur' option to image variants, configurable via API, to enable blurred previews for private images. Provided a Cloudflare Worker example for serving images from custom domains and announced an upcoming native feature for custom domain image delivery. Launched new bundled pricing for Cloudflare Images and Stream.
  3. Optimizing images on the webThis post introduces Cloudflare's Image Optimization Testing Tool (now unavailable) to help users identify image optimization opportunities. It details how image optimization impacts Core Web Vitals (LCP and CLS) and explains two primary methods for optimizing image dimensions: using `srcset` and `sizes` attributes, and leveraging Client Hints. It also highlights the benefits of newer image formats like AVIF and discusses the importance of setting appropriate image quality levels and including dimensions in markup to prevent Cumulative Layout Shift.
  4. Building Cloudflare Images in Rust and Cloudflare WorkersThis post details the implementation of Cloudflare Images and Image Resizing using Rust and Cloudflare Workers. It covers the reuse of Rust libraries across Image Resizing, Polish, and Cloudflare Images, the architecture of Image Resizing running at the edge and integrated with the cache, and the components of Cloudflare Images including core service, image resizing, delivery worker, and storage. It also explains image storage mechanisms (metadata, variants, original images), cache invalidation strategies (purge by URL and tag), and access restriction via signed URLs with expiration. Direct creator uploads are supported using Workers KV. Future work includes resizing hints on upload and serving images from custom domains.
  5. How Cloudflare Images can make your life easierThis post introduces Cloudflare Images as a solution to customer pain points related to image pipelines: egress fees, scattered buckets, slow load times, and high maintenance. Cloudflare Images offers zero egress costs, consolidates storage and variants, minimizes engineering effort through APIs and a dashboard UI, and processes images on the fly by automatically serving optimized versions based on browser support. It allows for generating up to 100 variants per upload and provides a one-time upload permission for users.
  7. AVIF image format supported by Cloudflare Image ResizingThis post introduces support for the AVIF image format within Cloudflare Image Resizing. It details the benefits of AVIF's superior compression over JPEG and WebP, its technical underpinnings (based on AV1 codec), and its adoption landscape. The implementation uses the rav1e Rust encoder. Caveats regarding encoding speed and lack of progressive rendering are discussed. The post provides examples of how to enable AVIF using Cloudflare Workers for dynamic format negotiation and using the `<picture>` element with URL-based resizing.
  8. Announcing Cloudflare Image Resizing: Simplifying Optimal Image DeliveryThis post introduces Cloudflare Image Resizing, a new product that allows customers to resize, crop, compress, and convert images to WebP. It leverages Cloudflare's edge network for global caching and can be integrated with Cloudflare Workers for dynamic image manipulation. The product aims to simplify optimal image delivery by allowing developers to store a single high-quality image and generate specialized variants on demand.
  9. NEON is the new black: fast JPEG optimization on ARM serverThis post details the optimization of the jpegtran library for ARM processors using NEON SIMD instructions. The author adapted previous SSE optimizations to NEON intrinsics, achieving a 2x speedup for the `encode_mcu_AC_refine` function and a 2.5x speedup overall for the jpegtran process on ARM, bringing performance closer to Intel Xeon processors.
  10. Doubling the speed of jpegtran with SIMDThis post details the optimization of the jpegtran tool for JPEG image compression by leveraging SIMD (Single Instruction Multiple Data) instructions. The author profiles jpegtran, identifies Huffman coding functions as performance bottlenecks, and applies SIMD intrinsics to accelerate loops for absolute value calculation, right shifting, and zero-value skipping. These optimizations resulted in a 2.25x speedup for the tested image.
  11. Improving PicoHTTPParser further with AVX2This post details the optimization of PicoHTTPParser using AVX2 instructions, moving from SSE4.2's PCMPESTRI. The optimization involves a shift from latency-bound to throughput-oriented logic by creating bitmaps of all token occurrences within a larger chunk of data (128 bytes) and combining masks for different token types. This approach, when compiled with gcc 4.9.2 and specific flags, resulted in a 1.68x to 1.79x performance improvement on Haswell processors.
  12. Experimenting with mozjpeg 2.0This post details the evaluation of mozjpeg 2.0 for Cloudflare's Polish product. It compares mozjpeg 2.0's compression performance and CPU usage against libjpeg-turbo using a sample of 10,000 JPEG images. The findings indicate that mozjpeg 2.0 achieves better compression ratios and successfully compresses a higher percentage of images, albeit with a longer processing time. Cloudflare has assigned an engineer to optimize mozjpeg 2.0 for Intel processors and plans to contribute improvements back to the project.
  13. Introducing Polish: Automatic Image OptimizationIntroduced Polish, an automatic image optimization feature. Polish offers two modes: Lossless (removes bloat like metadata, ~21% reduction) and Lossy (removes bloat and applies compression, ~48% reduction). The feature is available for paid accounts and operates by optimizing images in the cache.