Reducing Deployment Overhead


Over the past weeks, I completed several tasks that are worth highlighting from a technical perspective. While progress was made across all major areas, this post focuses specifically on efficiency, simplicity, and deployment overhead.

Compression Strategy: ZStandard

I revisited deployment compression. As described in a previous post, I initially chose LZMA2. While effective, it proved to be CPU-intensive and slow.

ZStandard Compression in Deployment

After benchmarking alternatives, ZStandard (.zstd) stood out immediately:

  • Archives are only a few percent larger
  • Compression speed is roughly 5× faster

This trade-off is clearly worth it. ZStandard is now the preferred and future-proof choice.

Architectural Cleanup: Landing Page Simplification

Next came a major simplification of the landing page. I removed a large amount of early-stage logic written before the overall architecture was fully clear.

Previous & New Landing Page

The long-running mascot has now been deprecated, along with several experimental features. As project complexity steadily increased, the landing page had become an unnecessary maintenance and failure concentration point.

Results:

  • +2000 lines of code removed
  • ~40 files deleted

This significantly reduced surface area and long-term maintenance cost.

Media Optimization: WebP & AVIF

Some deployments were still taking far too long. The root cause was media size.

All images were previously shipped as high-resolution PNGs — great quality, poor efficiency. For the Azure Web deployment alone, image assets had grown to ~300 MB.

WebP & Avif Compression

After evaluation, I switched to:

  • WebPfor standard-quality images
  • AVIF for high-quality images

Initial conversion — especially AVIF — was slow, even on a multi-threaded, high-performance system. However, the payoff was substantial:

  • No visible quality loss
  • Minimal to non-existent artifacts
  • Total size reduced from ~400 MB to ~100 MB

A 4x reduction with negligible downsides. WebP delivered very similar results, and I chose it for standard resolution to ensure the broadest possible compatibility going forward.

comments powered by Disqus