October 2012 Web Server Survey

In the October 2012 survey we received responses from 620,480,777 sites, an increase of 350K sites since last month's survey.

In spite of this, all major web server vendors lost hostnames this month with the exception of Microsoft – gaining around 3.5M – with a 0.58% rise in market share. This continues Apache's decline in market share, with a drop of 0.49 percentage points from last month.

In terms of active sites, all of the major vendors made losses. In terms of market share, both Microsoft and Apache made small gains, 0.5 percentage points for Apache compared with 0.1% for Microsoft.

In the top million busiest sites, nginx made modest gains in market share. Microsoft, Apache and Google all lost sites. This continues the downward trend on from last month for both Apache and Google, with Apache falling further back from the 60% threshold that it hit in August.

Although Apache dominates the overall market, where it is used by 58% of all websites, the secure server market paints a very different picture. Netcraft's SSL Survey found 2.3 million distinct, valid third-party certificates being used by HTTPS websites in October. Apache and Microsoft are almost neck and neck with 41.6% and 40.8% of the secure market share respectively, with the latter being noticeably higher than Microsoft's 15.7% share amongst HTTP sites.

Usage of nginx is also significantly different between the HTTPS and HTTP markets: although it is used by 11.9% of all sites in the Web Server Survey, it is used to serve only 2.3% of SSL certificates.

This month 516k hostnames moved from FC2 to Amazon, contributing to a net gain of 342k. FC2 is a provider of free ad-supported blogs and other web-based blogging tools. However, as many of these blogs are inactive, this change did not lead to an overall increase in the number of active sites hosted by Amazon.

Nasdaq has announced that it plans to make use of Amazon Web Services (AWS) to power its new financial data management system, which retains data required for financial regulation. This can be seen as a sign that large organisations are becoming more willing to make use of third party cloud offerings to host their data systems, and that the benefits associated with cloud-based solutions (low cost and high scalability) are increasingly outweighing the risks (security and accountability concerns). This follows Amazon becoming the largest hosting location last month. The quantity of web-facing computers owned by Amazon increased a further 5% this month.

Developer September 2012 Percent October 2012 Percent Change
Apache 362,714,083 58.49% 359,875,516 58.00% -0.49
Microsoft 97,368,803 15.70% 101,005,285 16.28% 0.58
nginx 73,976,009 11.93% 73,243,944 11.80% -0.12
Google 21,576,233 3.48% 20,947,340 3.38% -0.10
Continue reading

Most Reliable Hosting Company Sites in September 2012

Rank Company site OS Outage
hh:mm:ss
Failed
Req%
DNS Connect First
byte
Total
1 Qube Managed Services Linux 0:00:00 0.003 0.196 0.096 0.194 0.194
2 Hosting 4 Less Linux 0:00:00 0.003 0.142 0.101 0.204 0.391
3 Kattare Internet Services Linux 0:00:00 0.007 0.186 0.064 0.129 0.263
4 New York Internet FreeBSD 0:00:00 0.007 0.172 0.079 0.160 0.486
5 XILO Communications Ltd. Linux 0:00:00 0.007 0.275 0.103 0.319 0.537
6 Datapipe FreeBSD 0:00:00 0.014 0.100 0.016 0.032 0.048
7 www.logicworks.net Linux 0:00:00 0.014 0.199 0.082 0.473 0.585
8 www.choopa.com Linux 0:00:00 0.014 0.207 0.088 0.180 0.245
9 ServInt Linux 0:00:00 0.014 0.362 0.091 0.185 0.338
10 www.netcetera.co.uk Windows Server 2008 0:00:00 0.014 0.083 0.102 0.206 0.512

See full table

Qube Managed Services was the most reliable hosting company in September, responding to 99.997% of all requests throughout the month. The London-based company specialises in VMware cloud services, colocation, backups, and also offers PCI-DSS compliant hosting on both virtual and dedicated platforms. Qube also has infrastructure in Zurich and New York, and was also the most reliable hosting company during August.

Hosting 4 Less was the second most reliable hosting company, responding to the same percentage of requests, but with a longer average connection time. Hosting 4 Less has been operating since 1998 and offers both dedicated and shared hosting on Linux or Windows platforms, all backed by a 99.9% uptime guarantee.

Linux was the most prevalent operating system amongst the 10 most reliable hosting companies in September; seven of these sites were hosted on Linux servers, including www.qubenet.net and www.hosting4less.com, while two used FreeBSD and one used Windows Server 2008.

Netcraft measures and makes available the response times of around forty leading hosting providers' sites. The performance measurements are made at fifteen minute intervals from separate points around the internet, and averages are calculated over the immediately preceding 24 hour period.

From a customer's point of view, the percentage of failed requests is more pertinent than outages on hosting companies' own sites, as this gives a pointer to reliability of routing, and this is why we choose to rank our table by fewest failed requests, rather than shortest periods of outage. In the event the number of failed requests are equal then sites are ranked by average connection times.

Information on the measurement process and current measurements is available.

Phishing Alerts for Certificate Authorities

The internet community has been taught that one of the key steps in protecting their personal information on the internet is to ensure that it is entered only over an encrypted connection, perhaps by looking for the lock symbol in the browser address bar or web addresses beginning with https://. As a result, phishing attacks which make use of SSL certificates are especially dangerous as most users associate the presence of a valid SSL certificate with an increased level of assurance. Such attacks erode the reputation of Certificate Authorities and SSL certificates.

While the majority of phishing attacks run over HTTP, a significant number run on sites for which SSL certificates have been issued. In July 2012 alone, Netcraft found phishing attacks using a total of 505 unique valid SSL certificates from widely trusted issuers.

Although in some cases certificates have been issued specifically for the purposes of phishing the more common case is where well intentioned, bona fide certificate owners find that they are unwittingly providing facilities for phishing because their site has been compromised by an attacker.

Having access to timely, professionally validated alerts when phishing attacks occur is operationally efficient and responsible for certificate authorities, as well as an important part of preserving their company's reputation. It gives post issuance information on troublesome certificates and domains of which the certificate authority might otherwise be blissfully unaware.

Phishing Alerts are also a very valuable service for certificate holders, for whom it may be the first notification of a serious problem, giving them an opportunity to engage the attacker and wrest back control of their site before more harm is done.

Netcraft produces a continuously updated phishing feed that is very widely used. At least three separate third party studies have found it to be the most comprehensive feed available. The feed is used in all the major web browsers and it is also licensed by many of the leading anti-virus, content filtering, web-hosting and domain registration companies.

Phishing sites are submitted to the feed by the Netcraft Toolbar community. Reporters range from individuals submitting phishing mails that they have personally received, to specialist security researchers and several of the largest banks and financial payment systems. All submissions are carefully validated before being added to the feed. Well over five million unique phishing sites have been detected and blocked by Netcraft's community to date [September 2012].

GlobalSign commenced providing this service to all of its certificate owners in August 2012 (press release), and in the first month of the service around 70 distinct certificate owners were alerted to phishing attacks on sites where their certificates were deployed.

More information:

Please contact us (sales@netcraft.com) for pricing or further details about any of our services.

September 2012 Web Server Survey

In the September 2012 survey we received responses from 620,132,319 sites, a decrease of 8M sites since last month's survey.

A large portion of this drop was caused by a large network of linkfarmed domains disappearing from under the .com TLD, causing Apache numbers to suffer the most, with a loss of 10M sites. This resulted in a small drop in Apache's market share to 58%. Google also saw losses of 1M sites, but both Microsoft and nginx gained, with 840k and 1.5M new domains respectively.

Server headers for IIS 8.0 – the latest version of Microsoft's server software – were returned by 1,723 sites this month. This is an increase of 1,445 sites (+519%) over the six months since the public beta release of Windows Server 2012 in April, which uses IIS 8.0 as its default web server. However, only twelve of the million busiest sites were found to be using the software, seven of which are within Microsoft's own iis.net.

Amazon reached a significant milestone this month, with its strong and continued growth in the web hosting market now making it the world's largest hosting location by number of web-facing computers. The previous leader was China Telecom, which now has 116k web-facing computers against Amazon's 118k.

Netcraft's hosting provider server count uses a set of heuristics to identify individual computers, regardless of how many web-facing IP addresses each may have, or how many websites they serve.

Amazon has nearly doubled its count of web-facing computers within the past year, and this growth does not look set to slow down any time soon. The majority of these computers are located in the US (77%) and Ireland (13%), although smaller numbers of servers have started popping up in other locations within the past year, including the Netherlands, Singapore, Brazil, and Japan.

Although Amazon has the largest number of web-facing computers, these are used to host a relatively modest sum of 6.8M websites. 2.9M of these sites are served by nginx, which is closely followed by 2.3M served by Apache. A further 410k are served by Polyvore Web Server, which is used by sites within the Polyvore fashion social-commerce network. Only 2.4% (163k) of the sites hosted at Amazon are running Microsoft IIS.

Although Amazon's scalable, pay-as-you-go EC2 service supports Microsoft Windows, Linux is by far the most popular operating system to be found amongst all of its web-facing computers, including those used by CloudFront and S3. Nearly 97% of Amazon's web-facing computers were running Linux during September's survey.





DeveloperAugust 2012PercentSeptember 2012PercentChange
Apache373,069,75159.39%362,714,08358.49%-0.90
Microsoft96,529,58615.37%97,368,80315.70%0.33
nginx72,429,97611.53%73,976,00911.93%0.40
Google22,561,8543.59%21,576,2333.48%-0.11
Continue reading

Minimum RSA public key lengths: guidelines or rules?

The length of an RSA public key gives an indication of the strength of the encryption — the shorter the public key is; the easier it is for an attacker to brute-force. An attacker, armed with a compromised private key derived from a short public key, would be able to decrypt both past and future SSL-secured connections if she were able to incept the encrypted traffic. She could also impersonate the organisation to which the SSL certificate was issued if she has the opportunity to manipulate DNS lookups. Both the CA/B Forum (a consortium of certificate authorities (CAs) and major browser vendors) and NIST [PDF] (the agency which publishes technical standards for US governmental departments) have recommended that sub-2048-bit RSA public keys be phased out by the end of 2013.

According to the CA/B Forum's own Baseline Requirements [PDF] — effective 1st July 2012 — member certificate authorities are required to reject a request to sign an RSA public key shorter than specified in the following table:

Certificate expiry date Minimum RSA public key length
On or before 31st December 2013 1024
After 31st December 2013 2048

Nevertheless, these key sizes are not guaranteed as several CA/B Forum members have issued several non-compliant SSL certificates since 1st July 2012. Trustwave, Symantec, KEYNECTIS, and TAIWAN-CA have all signed certificates which fall foul of their organisation's requirement of 2048-bit RSA public keys for certificates expiring after 2013, demonstrating that the key length requirement is being treated as a guideline (which by definition is neither binding nor enforced), rather than a rule.

They are by no means the only CAs signing short RSA public keys: more than 10 years after Netcraft's first blog post on the topic and 12 years after RSA-155 [PDF], 512-bit RSA public keys are still appearing in SSL certificates. A 512-bit RSA public key was signed as recently as July 2012 by Swisscom.

Most, but not all, of the major browser and operating system vendors either disallow access or display a warning message when accessing a website using an SSL certificate with a 512-bit RSA public key. The latest versions of Safari (although not the mobile version on iOS 5.1), Opera, Google Chrome, and Internet Explorer (via an update to Windows; planned to be rolled out in October 2012). Notably, Mozilla Firefox does not yet reject such certificates.

Most Reliable Hosting Company Sites in August 2012

Rank Company site OS Outage
hh:mm:ss
Failed
Req%
DNS Connect First
byte
Total
1 Qube Managed Services Linux 0:00:00 0.003 0.170 0.133 0.270 0.271
2 Datapipe FreeBSD 0:00:00 0.007 0.079 0.017 0.036 0.054
3 iWeb Technologies Linux 0:00:00 0.007 0.127 0.085 0.170 0.170
4 www.netcetera.co.uk Windows Server 2008 0:00:00 0.007 0.045 0.139 0.281 0.701
5 XILO Communications Ltd. Linux 0:00:00 0.007 0.376 0.141 0.406 0.687
6 ReliableServers.com Linux 0:00:00 0.010 0.172 0.073 0.180 0.219
7 New York Internet FreeBSD 0:00:00 0.013 0.150 0.071 0.223 0.500
8 INetU Windows Server 2008 0:00:00 0.013 0.105 0.093 0.271 0.549
9 Server Intellect Windows Server 2008 0:00:00 0.017 0.149 0.055 0.111 0.278
10 www.choopa.com Linux 0:00:00 0.020 0.213 0.089 0.183 0.261

See full table

Qube Managed Services, a London-based company which specialises in virtual hosting, was the most reliable hosting company in August. With infrastructure in London, Zurich and New York, they provide Virtual Data Centres and Managed Hosting in addition to server co-location. Having placed in the top 10 twice before in the past six months, Qube's top place this month is well-deserved with 99.997% of requests succeeding.

Datapipe and iWeb Technologies placed second and third this month with the same percentage of successful requests. Datapipe, whose connection times were on average faster, runs its data centres entirely on renewable energy and are recognised by the Environmental Protection Agency as a Green Power partner. iWeb offers dedicated and virtual servers in addition to a number of advanced solutions such as Web Clusters, ideal for customers requiring high availability.

Netcetera and XILO also both placed highly this month with the same number of successful requests as Datapipe and iWeb, putting them in fourth and fifth places respectively.

As we've seen in previous months, the most reliable hosting companies are all powered by Linux, FreeBSD or Windows Server 2008. The last time a different operating system placed in the top 10 was when Windows Server 2003 appeared, back in February.

Netcraft measures and makes available the response times of around forty leading hosting providers' sites. The performance measurements are made at fifteen minute intervals from separate points around the internet, and averages are calculated over the immediately preceding 24 hour period.

From a customer's point of view, the percentage of failed requests is more pertinent than outages on hosting companies' own sites, as this gives a pointer to reliability of routing, and this is why we choose to rank our table by fewest failed requests, rather than shortest periods of outage. In the event the number of failed requests are equal then sites are ranked by average connection times.

Information on the measurement process and current measurements is available.