Which Web Server Is "Winning"

The Clash of Claims: Netcraft vs Port80

In November 2003, two well‑known industry surveys painted almost contradictory portraits of the web‑server landscape. The UK‑based Internet services firm Netcraft released a headline that Apache had captured a “significant percentage gain” over Microsoft’s Internet Information Services (IIS), tipping the global market share past the two‑thirds mark. Just days later, Port80 Software’s own Top 1000 Corporations Web Server Survey countered with the claim that Microsoft IIS still dominates the corporate segment, holding 53.8 percent of the market. The headlines made headlines, but the numbers behind them raised more questions than they answered.

To understand why the two reports diverge, it is essential to look beyond the raw percentages and examine the methodology each company used to arrive at its figures. Netcraft’s survey casts a wide net across the entire public internet, counting every hostname that responds to an HTTP request. Port80, on the other hand, focuses narrowly on the live, high‑traffic web sites of Fortune 1000 companies. The difference in sample size, scope, and the types of sites included explains much of the apparent discrepancy. Yet even within each methodology, there are further nuances - such as how parked domains are treated or how data is corrected for duplication - that influence the final numbers.

Both surveys rely on header analysis to detect the server software. When a browser requests a page, the server replies with an HTTP header that typically contains a “Server” field, e.g., “Server: Apache/2.4.41 (Ubuntu)” or “Server: Microsoft-IIS/10.0”. The two firms scan thousands of responses and then aggregate the data. Where the methodology differs is in which hostnames they decide to count, how they filter out duplicates, and how they handle the many domains that simply exist on a server but do not host active content. The way a survey handles these edge cases can shift the balance dramatically, as seen in the Netcraft case where a single parking service accounted for over 1.4 million server responses.

Netcraft’s claim that Apache now leads with over two‑thirds of the global web‑server market has been widely quoted in press releases and industry articles. That figure, however, derives from an uncorrected tally of 44.9 million hostnames. When the company applies its logarithmic adjustment to discount parked domains, the number of “active” sites drops to less than 20 million, and the percentage advantage for Apache shrinks. Port80’s report, based on 1 000 verified, high‑volume corporate sites, shows a very different picture: Microsoft’s IIS holds the majority. The contrast highlights how a single statistic can be misleading if the underlying data set is not transparent or if the adjustment logic is opaque.

The debate between the two surveys is more than a numbers game; it speaks to the broader question of what we mean by “market share.” Is it the proportion of hostname responses across the public internet? Or is it the fraction of corporate sites chosen by the largest enterprises? The answer depends on the audience. Web developers and small‑business owners might prioritize Netcraft’s broader view, while CIOs and enterprise architects may look to Port80 for signals that inform technology investment decisions. Understanding the distinction between these two lenses is the first step toward interpreting the data correctly and applying it to real‑world choices.

Unpacking Netcraft’s Methodology

Netcraft’s public survey lists almost 45 million hostnames that responded to HTTP probes during the November 2003 period. Of those, 30.3 million were identified as running Apache, and 9.4 million as running IIS. On the surface, the numbers appear clean: Apache dominates by a wide margin. But the key lies in how the survey interprets and filters that raw data.

The survey’s creators admit that the early days of the web made hostname counts a reasonable proxy for active sites, but the landscape has changed. Today, many hostnames belong to parking services or bulk‑hosting providers that run thousands of domains on a single server, often with identical software. In 2000, Netcraft described a “logarithmic correction” to account for such parked domains, reducing the number of active sites reported for the parking service futuresite.register.com from 1.4 million to 515. This method, however, still counts that one server’s 515 sites, effectively inflating Apache’s presence if a bulk host chooses Apache over IIS.

The biggest single swing in the November 2003 results came from Register.com, a parking service that shifted from Apache to a mix of Apache and IIS. Netcraft’s uncorrected data reflected a 2.8 percent increase for Apache, while the corrected numbers showed only a 1.25 percent gain. The fact that the headline relied on the uncorrected figure, while the body of the report presented the corrected numbers, suggests a potential disconnect between the message and the data. The correction itself depends on a mathematical model that assigns diminishing weight to successive domain counts on a server, but the model’s assumptions are not universally agreed upon.

Beyond parked domains, there are other sources of bias. Netcraft’s method does not distinguish between an active corporate site and a static marketing page that never receives traffic. A single large corporate server may host hundreds of domains, each counted separately, thereby skewing the overall market share in favor of the server software chosen by that host. Additionally, the survey relies on the “Server” header, which some administrators intentionally obscure or randomize for security reasons. If a large number of corporate servers mask their header, the data may underrepresent the actual usage of IIS or Apache.

While Netcraft’s figures are often cited as the industry benchmark, the methodology’s limitations become apparent when compared to Port80’s targeted approach. Netcraft’s wide net captures a broad slice of the public internet, but it does so with a resolution that is too coarse to discern the nuanced preferences of large enterprises. The survey’s claims of a two‑thirds share for Apache must therefore be contextualized: they reflect a raw count of hostnames, not necessarily the depth of investment, performance, or security posture that a company might consider when choosing server software.

Port80’s Targeted Survey Reveals Corporate Preferences

Port80 Software’s Top 1000 Corporations Web Server Survey takes a very different tack. Instead of sampling every hostname, the survey examines the live sites of the Fortune 1000, a list of the largest companies by revenue. By focusing on high‑traffic, high‑importance sites, Port80 seeks to gauge which server software those enterprises are actually deploying, rather than merely which software appears most frequently across the entire web.

The methodology remains similar in that each hostname is interrogated for its HTTP “Server” header, but the sampling frame is far narrower. The result is a cleaner, more reliable dataset: 1 000 unique, high‑traffic sites that can be individually verified for authenticity. Think of it as the difference between a fishery that casts a huge net and a fisher who uses a fine‑mesh line. The latter approach yields only the most valuable catches - large, well‑maintained corporate sites - while the former collects everything in the area, including debris and small fish that add little value to the analysis.

In the 2003 snapshot, Port80 reported that Microsoft IIS commanded 53.8 percent of the Fortune 1000 market, with Apache at 15.4 percent. These numbers are driven by well‑known corporate sites such as Walt Disney, BankOne, and the Gap, which all run IIS. The survey also notes a trend: many of these firms are upgrading from older Windows NT platforms to Windows 2000 and Windows Server 2003, indicating a long‑term commitment to the Microsoft ecosystem. The data thus reveal that large enterprises favor IIS, likely due to tighter integration with other Microsoft products, easier administration on Windows, and the support for the .NET framework.

The contrast with Netcraft’s headline becomes clear when you consider that the corporate environment places different demands on a web server than a small hosting company or a personal blog. Large enterprises need to handle massive traffic spikes, secure multiple applications, and maintain compliance with stringent security standards. Port80’s survey shows that the majority of Fortune 1000 sites are built on a platform that satisfies those needs, while Apache, although popular among small and medium‑sized sites, is less prevalent in this high‑stakes segment.

It is worth noting that Port80’s data are publicly available for free, and the survey is updated monthly. Anyone can view the breakdown at Port80’s survey page. The company also publishes additional resources, such as masking techniques for server headers, which can be found under the “Related” section linked in the main article.

Beyond these two primary sources, the industry community often publishes supplementary studies, including SSL‑specific surveys and vendor‑agnostic reports on server performance. While the numbers from Netcraft and Port80 are frequently cited, a well‑informed decision should incorporate multiple viewpoints, recent security advisories, and an assessment of the organization’s unique operational context.

Author Bio
Joseph Lima
COO and Director of Product Development, Port80 Software

Joe Lima has led product development and support teams at Port80 Software since its founding. He brings experience from a range of internet, wireless, and software companies, focusing on server‑centric technologies. As a lecturer at the University of California, San Diego and a published author on web‑server technologies, Joe Lima combines deep technical knowledge with a practical understanding of the challenges faced by web administrators.