Research: Comparing the performance of IP routing and onion routing (Tor)
The world is currently witnessing many political conflicts. Repressive governments are imposing sanctions on free speech. Many of these governments block the general public from accessing some websites, which led people in these countries to find solutions to be able to browse the internet freely. Researching solutions to this problem led to the practical deployment of a prototype model for a system to unblock internet access based on a network structure relying on onion routing (TOR). Tor has been developed to offer free internet access to everyone, while inflicting no harm to anyone.
A recently published paper evaluated the performance of the TOR network in comparison with direct access via IP routing. The research study used the average response time for downloading web pages with various sizes in KB as a metric for comparison. The results of this study showed that the difference between the response time of direct access (IP routing) and access via TOR varies as per the size of downloaded files. The average latency period induced by relays across the TOR network was found to be 0.72 seconds. The difference of average response time between the two means of internet access decreases with the file size exponentially, to be less than 7.6% when the file size is greater than 124 kB, The reason for this is that the delay for routing traffic through the relay nodes of the TOR network has more effect on the first data packet of the same connection, but the rest of the packets come in succession with negligible effect on reduction.
The measurements were made via Mozilla Firefox on Windows OS. The browser’s cache was eliminated via its deactivation in the settings of the browser, and also by erasing it manually after each measurement was made. This was essential because even though cache was deactivated, somehow some data had still been cached.
The measurements included measuring the time needed for completely downloading a web page, or saving a file to the hard disk. No applications, apart from Firefox, were running in the background on the OS, except the antivirus system, all through the measurement process.
Comparison was made using www web pages and files, due to the fact that www represents a significant share of the total internet traffic. Download times for www pages and files were measured in sets with five rounds each. Measurements were done in five periods during a single day. The PC used in the experiment was connected to an ISP network via a small LAN with a bandwidth of 6 Mbps for download speed and 2 Mbps for upload speed.
The following 6 modes were used during measurements:
1- Normal operation of the PC without Tor.
2- Same as mode 1, with onion router (OR) via 0.5 Mbps in LAN.
3- Connecting via Tor using its default settings.
4- Using manually configured Tor, with fixed ingress and exit ORs.
5- Using manually configured Tor, with ingress and exit ORs picked up from a manually configured list of 15 ORs (this mode was used during speed measurement of www pages download).
6- Using manually configured Tor, with ingress and exit ORs in the test network (this mode was used during speed measurement of www file downloads).
For all modes using Tor for active internet traffic (modes 3 to 6), the network of ORs included 3 ORs. In mode No. 4, the OR network persistence time was manually set to 30 minutes (default 1 min.) to allow keeping the same network conditions (the same network of ORs) for the complete set of measurements.
In mode 6 there were two own experimental ORs used: the ingress OR was located in another site apart from the client’s network, the exit OR was in the same network where the client was located. It means that the only OR that could vary was the second OR. As the Tor network is subject to relatively frequent changes, this was taken into account in evaluation of the results of the experiment. Measurements that were associated with significant change occurring during download causing longer delay time were excluded from further processing. The number of such measurements was not significant and its influence on the total results was almost negligible anyway.
Measurements taken throughout the experiment of the study showed that connecting to the internet via Tor is significantly slower when connecting normally via IP routing. The delay caused by the default Tor settings can be reduced greatly, yet still remains significant. As such, one can conclude that the Tor browser offers a convenient tool for anonymizing internet traffic, yet its use is better to be limited to very special use cases, due to the fact that the price (i.e. the delay caused when connecting to the internet) is significantly high.