HAProxy

The Reliable, High Performance TCP/HTTP Load Balancer

Quick links

They use it !

Some happy users want to report their experience with HAProxy. Some sites openly confess they use it. Some users are simply curious about who currently uses it. So I'm assembling all these information here for them, in alphabetical site name order. If you think something is inaccurate, or if you'd like to send a few sentences about your experience and have a link to your site, please contact me.

Quoting A0 Labs' CEO Olivier Warin: “A0 Labs is a French hosting company specialized in critical applications which need high performance and security. We have been using Haproxy for several years now and deploy it for our customers to provide high availability and performance to their web sites”. I would personally like to add that A0 Labs helped a lot on the FreeBSD port a few years ago, regularly helps by benchmarking and testing new releases, and contributes to quite a number of other opensource projects, not just HAProxy !

Border Stylo is one of the companies making heavy use of the stunnel + haproxy tandem on Amazon instances. Carlo Flores, Sr Operations Engineer, says : “Border Stylo is a social networking site, and we use HAProxy at the core of our stack after benchmarking scale (speed, KB/connected user) vs Pound and Nginx for our architecture serving www.writeonglass.com and retrollectapp.com. Operations Engineers here consider the mailing list one of the better resources on the net, if only for Willy and Cyril's musings and configuration files, and contribute what we can with our ctl script at https://github.com/borderstylo/haproxyctl. Personally, I think the statistics generation and halog sources is at least as impressive as HAProxy's speed and resource management”.

An article on High Scalability gives some details about the architecture behind DISQUS and the traffic it has to deal with. In short, 17000 requests per second, 250 million visitors. They're using HAProxy both in front of the web servers and in front of the database. A lot more info are found in this detailed presentation.

Egnyte is a Cloud File Server. Quoting Sachin Shetty : “Haproxy is a fantastic feature-rich load balancer and we at Egnyte have been using it for a while. Apart from using haproxy for standard application load balancing, we are using haproxy to overcome some limitations of Apache like use queue timeout to prevent backlogging in Apache when application servers are loaded. We also use haproxy for load direction to route requests i.e. send requests to specific server under specific conditions and failover accordingly”. Thanks guys, your nice feedback is much appreciated !

Hctor Paz, the sites' sysadmin, reported on the HAProxy mailing list : “We use haproxy to handle web traffic for peruvian news sites: elcomercio.pe, peru21.pe, etc. Around 2k session rate in peak hours. Haproxy is the most reliable part of our architecture”.

Farmville is one of the most popular online games, edited by Zynga. Mark William indicated here that mid-2010 the site had over 70 million active users a month. While Zynga doesn't explicitly advertise use of HAProxy, they don't hide it either as they report using RightScale at Amazon EC2 to scale seamlessly, and even the error page has it in its URL.

Fedora is the community-driven distribution behind Red Hat. The Wiki explains how HAProxy is used there, and even provides links to the stats pages.

Free is a major player among the french ISPs. Free has always promoted the use of free software, and has been using HAProxy for many years. The Webmail and the file exchange service have been the most heavily loaded deployment ever reported in terms of network bandwidth, with more than 5 Gbps of traffic at any moment. They're used to provide extremely valuable feedback which has contributed to make the 10Gbps performance something real and to get TCP splicing a reliable solution.

There's a lengthy article on the Github Blog entitled "How We Made GitHub Fast". It explains in depth how the GitHub architecture works, and there's a lot to learn there for anyone who's planning on starting a scalable site. Interestingly, a second article here gives a few more details as to why they're not only using HAProxy but Ldirectord too (eg: smaller memory usage in VMs).

The Imgur guy describes his architecure choices on Reddit here and why HAProxy makes a good choice for him here. The full thread is quite informative about what issues such fast-growing sites are facing.

Jon Watte, IMVU's CTO, describes on slideshare how IMVU's architecture works and how it scales. Haproxy is just one small piece in the puzzle there despite being on the front. The site's home page indicates the number of concurrent users in real time (more than 120k when last checked). It's nice to see some large sites sensibilized to latency and report their usage numbers.

ITA Software is one of the companies who acknowledge use of Open-source components. Their experience is best described with their own words : “ITA Software is a leading provider of innovative solutions for the travel industry. Haproxy's power, flexibility, and reliability, have quickly made it a valuable part of ITA's infrastructure that supports leading travel compaines worldwide including American Airlines, Bing, Continental Airlines, Kayak, Orbitz, Southwest Airlines, United Airlines, US Airways, Virgin Atlantic Airways, and others”.

The Olark guys explained here how they set up their site with high availability, and some of their decisions to ensure uninterrupted service in case something goes wrong. They give a bit more details about the monitoring and some architecture fixes here.

pfSense is an open source firewall based on FreeBSD and has an haproxy optional module along with a web interface for configuring haproxy. More information on the package is available here.

Playfish, now part of EA, offers a wide number of online social games. As of 2010, the site already counts 10M daily users and 50M monthly users. As a number of such other gaming sites, it's hosted in Amazon EC2 and incoming traffic is load balanced using HAProxy.

Ravelry is a social network dedicated to knitting that was founded by Casey and Jessica Forbes in 2007. It was quikly welcomed with a great success and Casey had to perform important changes several times in the architecture to follow the growth. He explains his adventures here. In 2008, one year after the project was born, Casey told me : “HAProxy is fantastic. We use it at www.ravelry.com to handle 5 million or so requests per day”. And now we're in 2011... Their project is quite original and I wish them a long success story !

It's probably the only site who is so open about its infrastructure that even their HAProxy configuration is available to everyone !

As described in their architecture overview, Red Hat uses HAProxy as the load balancing solution in its cloud architecture OpenShift. While I know for sure they're not the first cloud provider to use it, I can say that they're the first one to openly admit it and that's nice from them (their architecture overview is well detailed and worth a read BTW).

Ben Timby of SmartFile says : “Hundreds of gigabytes of data flows from SmartFile through HAProxy each day. SmartFile uses HAProxy both for HTTP and FTP protocol load balancing. The PROXY protocol makes it possible to provide highly available and lightning fast FTP service. Without the many features of HAProxy and the support of Willy Tarreau dragging the old FTP protocol into the 21st century would have been near impossible. Many thanks for such a stable and flexible product”. Please note that SmartFile was kind enough to fund development of the server-side PROXY protocol implementation.

The same team is managing both sites. They're well known and have high expectations on reliability and quality of service. They've funded the development of the anti-abuse features in HAProxy.

Jeff Atwood says : “We're big fans of HAProxy, which the guys at Reddit turned us on to. It has been working flawlessly for us in load balancing Stack Overflow between two - and now three - servers”.

Transloadit is a file upload service for web applications. One of its co-founders, Kevin van Zonneveld, explains here that he uses it for the content switching, and also gives some hints about setting up logging under Ubuntu. I'll probably have to put that into the doc because it looks like it was not obvious.

As of April 2012, TubeMogul is the biggest Real Time Bidding video ads platform. Nicolas Brousse, Lead Operations Engineer, says : “We use Haproxy in four different EC2 regions and five Availability Zones. It allows us to handle over 10 billion HTTP bid requests a day and deliver over six billion videos ad streams last year”.

As of December 2011, Tuenti is the most trafficked website in Spain with more than 12M users and 40 billion page views a month. In the following presentation, Senio systems Engineer Ricardo Bartolom explains the previous load balancing infrastructure, the neew load balancing strategy, as well as the reasons why they have chosen HAProxy as the Layer7 load balancing solution : www.slideshare.net/ricbartm/load-balancing-at-tuenti.

This article gives some details on the Tumblr architecture. As of Feb 2012, it's at 500M page views a day, 40k requests/s with plans to go to 400k, and observes a 30% monthly growth. It involves more than 1000 servers, and employs 25 HAProxy, 15 Varnish and 8 Nginx to make this run smoothly, the same winning trio that is found on many large sites !

John Adams explains here how they scaled Twitter to support a traffic growth of 1358% in 2009. It looks like they adopted the principle of "one component per function" which generally scales the best. There is very little information about the load balancing part in this slide show, but it also happens that scaling the rest is much more important.

There was a presentation from Virgin America at LinuxCon 2010 where they explained how they migrated to full open-source. Among the numerous products involved, HAProxy is used for the load balancing. The complete presentation is available in PDF format here. A quick summary of the presentation is also available.

The W3C obviously doesn't have to be presented to you if you're working in web environments. Yes, when you visit the W3C, you're passing via HAProxy as is explained here. From some past discussions, I remember it also helps protecting the whole site against unintentional misuse caused by excess of document validation.

This probably is the fastest site I've ever seen and certainly one of the most highly stressed I know. They deliver the nice counter you can see on the HAProxy page to millions of web sites around the world. To get an idea of the load, consider that each time one of these sites' page is viewed, they receive a request. Their response time and availability are obviously critical to those sites, and they excel in this area with sub-millisecond response times. This site perfectly fits HAProxy's strengths, and some of the high performance optimizations directly come from their feedback.

As described in this highscalability article (slides here), YouPorn stacks HAProxy, Varnish and Nginx to achieve 300000 requests per second and 100 Gbps of traffic, all of this producing 15 GB of logs per hour. With numbers 10 times larger than any of the other sites listed here, it looks like porn will always be most of the net's traffic !

Contacts

Feel free to contact me at for any questions or comments :

• mailing-list :
  Read the archives on Formilux site
• e-mail :

Some people regularly ask if it is possible to send donations, so I have set up a Paypal account for this. Click here if you want to donate.