Download websites that use Apache Traffic Server

About Apache Traffic Server

Almost every technology in a website profile was chosen by whoever built the website. Apache Traffic Server usually was not. The proxy sits in front of many sites at once, installed by a hosting company, a telecom operator or a media platform, and the individual domain behind it may have no idea the software exists.

The distinction changes how a count should be read. When we record the proxy on 56,154 sites as of October 2026, the underlying population is not that many independent decisions. A single deployment can account for a very large block of domains, so the figure measures reach rather than adoption, and anyone counting customers from it will overcount badly. The dataset is still the right instrument, but the question it answers is which domains sit behind the proxy, not how many organisations chose to run one.

The software came out of a commercial caching product, was donated to Apache Software Foundation in 2009 and has been developed in the open under Apache 2.0 ever since. Written in C++, Apache Traffic Server is built for the sort of throughput that makes commercial delivery contracts expensive, and the operators who run it are the ones for whom bandwidth is a line item rather than an invoice they forward to a client. A plugin architecture means large deployments are frequently customised past the point where stock behaviour is recognisable.

Version data is unusually good for infrastructure of this kind. We observe 30 distinct release strings, with 9.2.9 recurring across a substantial part of the deployments we can identify, which reflects an operator population that patches on a schedule rather than a long tail of forgotten installs. A cache that terminates TLS for a large estate gets maintained, because the people running it are the same people who get paged when it stops.

United States accounts for 40,367 of the domains we attribute to the proxy, and .com runs through the list, both of which follow from where the large hosting and media platforms are incorporated rather than from where the software is favoured. Detection depends on response headers, and headers are exactly what a careful operator strips. Server identification can be removed, rewritten or masked by another layer in front, so the count is a floor and the most deliberately configured deployments are the ones least likely to be in it.

The people who need this file are selling to operators, not to websites. Hardware, transit and bandwidth vendors, observability and security companies, and competing edge platforms all want the set of domains resolving to a self operated cache, because the presence of one is evidence of an in house infrastructure team with a budget and a reason to talk. Reading one domain at a time cannot show you the clusters, and the clusters are the product. Grouping the list by network, by suffix and by co-occurring origin software turns a long column of hostnames into a short list of accounts worth calling.