On the Accuracy of Tor Bandwidth Estimation

Abstract

The Tor network estimates its relays’ bandwidths using relay self-measurements of client traffic speeds. These estimates largely determine how existing traffic load is balanced across relays, and they are used to evaluate the network’s capacity to handle future traffic load increases. Thus, their accuracy is important to optimize Tor’s performance and strategize for growth. However, their accuracy has never been measured. We investigate the accuracy of Tor’s capacity estimation with an analysis of public network data and an active experiment run over the entire live network. Our results suggest that the bandwidth estimates underestimate the total network capacity by at least 50% and that the errors are larger for high-bandwidth and low-uptime relays. Our work suggests that improving Tor’s bandwidth measurement system could improve the network’s performance and better inform plans to handle future growth.

Appendix

Figure 7 shows for each relay position the uptime by absolute discovered capacity quartiles. We observe that for all positions relays with higher discovered capacity have higher uptime, although we notice that guard relays have higher overall uptime (due to the additional uptime and stability requirements to get the Guard flag), and middles have a larger number of relays with low uptime.

Fig. 7.

The effect of relay uptime, where relays are ranked by their absolute discovered capacity. (a) Shows exit relays split into 4 sets by rank. Summary of the absolute discovered capacities (in Mbit/s): min = $-88.0$, Q1 0.00, med = 0.0, Q3 = 39.1, max = 707. (b) Shows guard relays split into 4 sets by rank. Summary of the absolute discovered capacities (in Mbit/s): min = $-90.3$, Q1 = 0.00, med = 0.428, Q3 = 31.1, max = 881. (c) Shows middle relays split into 4 sets by rank. Summary of the absolute discovered capacities (in Mbit/s): min = $-169$, Q1 = 0.00, med = 0.00, Q3 = 5.13, max = 774.

Figure 8 shows the capacity of guard and middle relays after the speed test by quartiles of relative discovered capacity. It includes only relays with at least 75% uptime. It shows that for high-uptime relays in both positions, most of the discovered capacity is among the largest relays. We can especially see that for middle relays, the low amount of discovered capacity is due to the large number of relays with very little total or discovered capacity.

Fig. 8.

Absolute capacity of relays after the speed test, where relays are ranked by their relative discovered capacity. Relative discovered capacity is computed as $(\text {after}-\text {before})/\text {after}$. Includes only relays with uptime of 75% (273 days) or more during the year preceding the speed test. (a) Shows guard relays split into 4 sets by rank (1,238 guards had at least 75% uptime). (b) Shows middle relays split into 4 sets by rank (983 middles had at least 75% uptime).

Figure 9 shows the effect of relay capacity on the discovered capacity by position when only relays with at least 75% uptime are considered.

Fig. 9.

Absolute capacity of relays after the speed test, where relays are ranked by their absolute discovered capacity. Absolute discovered capacity is computed as $\text {after}-\text {before}$ Includes only relays with uptime of 75% (273 days) or more during the year preceding the speed test. (a) Shows exit relays split into 4 sets by rank (379 exits had at least 75% uptime). (b) Shows guard relays split into 4 sets by rank (1,238 guards had at least 75% uptime). (c) Shows middle relays split into 4 sets by rank (983 middles had at least 75% uptime).