Raph Levien (raph@kiwi.cs.berkeley.edu) says:
I believe that I am providing a useful service with my remailer list, but I have received one complaint about the frequency of pinging.
If I may suggest without implementing :-) use adaptive pinging. There is little point in sending one ping per hour while getting no answer, then getting 24 answers all at once, and then doing the same thing all over again the next day... A more general way to put this is that pinging much more frequently than necessary does not give any more info. If you measure up time and latency both in days, there is no need to ping more than once a day. If you measure up time and latency respectively in days and minutes, there is still no need for very frequent pinging. The only case where you need frequent pinging is when a site keeps going up and down and you want to distinguish latency due to down periods from operating latency.... But from a practical standpoint, these need not really be distinguished. Most mailers are configured correctly to retry failing connections, so that connections that are down only for a few hours are seen only as long latency. In conclusion: Start with pings at random phase, and 180 minute period. If for the last 16 pings of average period n minutes (random phase), the average latency is more than 4xn minutes, triple the period. Use a maximum period of once a day. If the average latency is less than n minutes, divide by three the period. Use a minimum period of one hour. You could do the same thing for very stable sites (stable latency): they require less pinging. Short latency stable sites need not be pinged every hour. The rule may have to be fixed a bit one way or the other, but it would help both your pinging script and slow remote sites (personal remailers) by cutting traffic for both. Pierre. pierre@shell.portal.com Still, there is no harm in making pinging traffic look more like real traffic. Like you said chaining is questionable for pinging (and if it comes from a remailer site, and goes back to a remailer site, it would not be obvious if it still met 1, 2 and 4):
1. Encrypt the ping, so it looks like so much PGP traffic.
2. Pad it with a random amount of junk (but not _too_ much :-), so traffic analysis based on size will fail.
3. Chain it through other remailers. A good approach might be to choose two random remailers out of the "top five," and sandwich rebma between them. The drawback is that it penalizes rebma for their latency and failure rate, but this might be acceptable.
4. Randomize the time that the ping is sent.