ghio@temp0099.myriad.ml.org (Matthew Ghio) wrote:
You could also use MD5 hashes of the messages, in which case a list of 1000 message-IDs would take only 16K. (In the unlikely event of a hash collision you could download those two messages seperately. Unless the number of messages was huge (millions), you could probably get away with using only a 32 or 64-bit hash function.)
Okay. Let's suppose that there are 10,000 messages (more realistic for a large remailer i think) And I am going to spread it over five servers, And I use a 32-bit hash function (one in four billion chance I get someone else's message) First I download the list of Message IDs/Hashes. (40,000 bytes) Then I download the recipient list. (another 40,000 bytes) I find a message for me. Let's suppose the messages are 20K each. I send each server a list of the messages I want (10,000 bits, which is 1,250 bytes each, so 6,250 bytes total) Finally, I get back five 20K messages from each of the five servers. So that's a total of 80K to download the IDs/recipients lists, 6.25K to upload the requests, and 100K to download the message pieces, to read my 20K email. I guess that could work.