Substack saved 90% on image processing infrastructure with imgproxy
Whether it’s cost savings or speed gains, it’s always insightful when our clients share statistics about using imgproxy in their infrastructure. In this post, we’ll reveal how Substack, who processes 5 million images daily, managed to reduce the average asset processing latency by 1.5X and slashed costs by 90%.
Substack is a newsletter subscription platform designed to empower independent writers, podcasters, videomakers, musicians, scientists, and culture-makers of all kinds by providing them with the tools to publish, distribute, and monetize their content seamlessly. They can easily start to shape and distribute their newsletters, get discovered on the web, and offer subscription plans so readers can directly support the creators.
Today Substack boasts more than 35 million active subscriptions, including 2 million paid subscriptions. The platform brings together many popular content makers across various topics: some are just getting started on Substack, but most are well-established. You can subscribe to (and even meet) writers like Margaret Atwood, George Saunders, Kareem Abdul-Jabbar, and Erik Hoel.
Ticking all the boxes
It would follow that a platform for writers would primarily involve a lot of text. But a picture is worth a thousand words, and illustrations are quite prevalent even here. In fact, Substack processes and transforms a gigantic volume of images — about 5 million per day. In terms of what they do with those pictures, Substack’s image operations mostly involve displaying, resizing and optimizing them.
Since the team uses pragmatic approaches to problem solving and ships high-quality products. They needed a stable, self-hostable and open source solution.
Despite the ease of use of their previous solution, imgix, the team felt they could find a more cost-effective option. They found the pricing for certain operations, like image resizing and optimizing to be higher than they’d prefer. So, they began exploring other solutions offering the same range of features at a price point that matched their budget.
imgproxy’s OSS nature and website immediately stood out to the team. Initially, the team thought about leveraging the open source version, but since they planned to use the advanced “smart centering/cropping” feature a lot, they decided to opt for Pro from the start.
The infrastructure of image processing
Their positive impressions were further confirmed once the project’s engineers started to work more closely with the imgproxy team who helped them with deployment. The platform is run on the popular AWS hosting, and it only took about 2 weeks to fully switch traffic over to the new imgproxy deployment.
Previously, this team was investing a substantial amount on image processing. However, after transitioning to imgproxy, they anticipate a significant reduction in costs–potentially up to 90%. Alongside the financial benefits, the average processing latency for their images has also improved, decreasing from 500ms with imgix to 350ms with imgproxy.
We were pretty happy with imgproxy from the moment we found it! The money savings is probably the biggest piece but it’s also impressive that imgproxy has not significantly increased our ongoing maintenance efforts compared to imgix.
Large numbers like this might seem a natural outcoming for high-load media or e-commerce projects that process millions of images a day, but we’re confident that digital projects from other industries and emerging startups can also benefit from making the switch to imgproxy! If you want to talk about how our product can help you optimize image processing in terms of speed and cost, drop us a line.