warms0x

on the internet

Home Contact On GitHub

Recent Posts

Scaling .onion sites

A while ago a question was asked on the tor-talk mailing list about “scaling a .onion site.”

First off, it’s pretty neat that enough people are using Tor Hidden Services such that scaling a site running a Hidden Service is required.

I figured I would write up a little guide on how to set up a “scalable hidden service” to help those looking to build genuinely interesting services out there.

Step 1

Actually setting up a Hidden Service for the first time is likely the biggest hurdle most people will face. Creating a useful Hidden Service is a much bigger challenge than creating a scalable one. In order to set up a Hidden Service I highly recommend this guide from the Tor project.

Step 2

In my opinion, the easiest way to scale a hidden web server is to add more hidden web servers. This can be accomplished with a load balancer) sitting in front of the web servers. In the case of a Hidden Service, you would simply point the Hidden Service at the load balancer instead of the web server itself. This can offer additional benefits in the way of checking the availability of the backend web server or even showing a helpful “sorry the site is down” web page.

I’ve gone ahead and created an example in a repository called scalable-onions of this technique. In the repository I’ve not included any of the details of setting up a Hidden Service, I’ll leave that up to the ready him/herself.

In this example we’ll set up a simple static file serving site with thttpd as the web server and haproxy as the load balancer. I’m also using Tails as my base operating system.

Packages to install:

  • haproxy (sudo apt-get install haproxy)
  • thttpd (sudo apt-get install thttpd)

Configuring thttpd is quite easy, the only changes I have made in my example thttpd.conf were to disable “chrooting” (which you should use!) and updating the paths to serve up a directory in my home directory. It is also worth nothing that the web server is bound to 127.0.0.1 which means it cannot receive connections from any other host other than the machine that is running the server itself.

Configuring haproxy is where the magic happens that helps make the .onion site scalable. In my example haproxy.cfg I only have one “backend server”, but in your set up you would likely have multiple backend servers which will allow you to serve up more content.

Note the listen statement on line 28, here again we’re only listening for connections on 127.0.0.1 and on port 8080. This means that the Tor Hidden Server configuration will be set up to connect to port 8080 on the local machine. I think this is a good security measure to undertake as it will ensure that only local connections on the host running Tor and HAProxy will work.

Another thing to note is that the traffic between HAProxy and the multiple backends will not be encrypted unless you are actually running web servers with SSL enabled. If you wish to do this you will need to look into running HAProxy with “TCP Load Balancing” instead of the currently configured “HTTP Load Balancing.”

For running a static site, this combination of HAProxy and multiple web servers should be enough to scale to additional demand.

Step 3

If you’re running MediaWiki or some other “dynamic” web application, then scaling the service will likely involve more than just adding more web servers.

In the case of MediaWiki, or other PHP-based applications, I suggest looking into tools like XCache or other PHP accelerators to serve each inbound request faster.

Other web applications, like those based off of Django or Rails will have their own needs or “optimization patterns” that you can follow to get efficiency and speed per request.


Conclusion

Due to the way that Hidden Services work, you will still need a single end-point for Tor to connect to, so using a load balancer is your best bet. There’s a lot that you can do to optimize a web application that is completely unrelated to Tor though, so before you shell out more money for another server, it’s worth learning more about what is exactly causing the slowness in your hidden service and addressing that.

Thanks RiseUp

I find myself appreciating the services that RiseUp provides more and more as the days pass by.

Today I discovered that my existing email account also gives me a Jabber (also known as XMPP) account under the same account name. I also discovered two new beta services:

  • RiseUp VPN - I wish I could summon the courage to use this service as I am sure it would be invaluable. Unfortunately I’m very cautious to never leak or connect to any of RiseUp’s services except through Tor. This is not due to a lack of trust of RiseUp, but a severe lack of trust of any bad actors between myself and RiseUp.
  • RiseUp Etherpad (also accessible under this .onion link).

The men and women behind RiseUp are really doing a bang-up job, I only wish I could do more to support them in their quest.

Until BitCoin takes off, or some form of anonymous cash mailing becomes viable/securable, I’ll have to continue to cheer from the sidelines.

Regardless, thank you RiseUp!