@Ivan

>Some best practices definitely would be awesome to have about running on common (embedded) hardware. Clear notification like "your Commodore 64 is to slow to be a good relay" would also be useful.

I agree about the need for guidelines but I disagree about the content of the guidelines that are needed. The data I see so far, including your report of a Pi with 7000 connections, is a clear indication that minimal hardware capabilities are NOT the guidelines that are needed (unless a relay with 7000 connections is still considered "harmful" or "useless").

My own Pi-based relay https://atlas.torproject.org/#details/707A9A3358E0D8653089AF32A097570A96400CC6 has just reached 1300 connections; CPU utilization: practically zero; memory utilization: 14.5%. The increase from 500 to 1300 connections required memory utilization increase of just 2%, from 12.5% to 14.5%. Clearly, hardware of the $35 Pi has absolutely nothing to do with residential relays being useful or not, save the (recently reported here) anomaly of an operator who has symmetric BW of 160 mbps to the home.

So guidelines on hardware are evidently not needed for "normal" residential ISP bandwidth: it has been amply demonstrated that even a dirt cheap Pi is not the bottleneck, no need to spend further effort on this until the REAL bottleneck is resolved: the network.

To continue the story, the above relay of mine with 1300 connections has consensus BW rating of 38 (thirty eight). Why? Who knows. I get zero feedback on the reason for this.

To further continue the story, my 2nd relay https://atlas.torproject.org/#details/31B8C4C4F1C78F923BD906769297B15A428C4A04 that currently has about the same Atlas-measured BW as the first relay (132 vs 153 KB/s) and is based on exactly the same hardware and software, is clinically dead with almost no connections and BW rating of 13. Why? Who knows.

What is needed is a standardized feedback on WHY the relay has such a low rating. This could cause at least part of the operators to take care of the bottleneck (eg moving the relay to another location, or abandoning the home relay and replacing it with a hosted one). And if the home relay is indeed as harmful as some people here think, the recommendation should be issued to shut it down, instead of leaving it hanging there doing nothing or even harming Tor. Such feedback could significantly improve the quality and effectiveness of Tor.

Based on the discussion here, the people who run Dirauths and bwauths know very well (or at least can easily find out) the reasons for relays getting low rating - why not automate the communication of the reasons to relay operators in clear, unequivocal and actionable terms? I get the feeling that people are trying to be "politically correct" here and it's a pity (although they DO respond fully and frankly when asked a direct question).

Rana