From: jim@bilbo.suite.com (Jim Miller)
Date: Thu, 12 May 94 12:13:09 PDT
To: cypherpunks@toad.com
Subject: Re: Message Havens
Message-ID: <9405121909.AA28859@bilbo.suite.com>
MIME-Version: 1.0
Content-Type: text/plain




> Downloading the whole message base to scan for one's
> messages will place a massive load on net.resources, and
> probably a prohibitive load on most people's terminals. 

> 


Here's an alternative to using tags that need to be agreed upon in  
advance:

I call it "Indexed Message Pools"

The key ideas:

Each message sent to the message pool will be encrypted in the  
recipient's public key. (nothing new here)  The Subject: line for the  
message will be the MD5 hash of the message body.

For each message sent, the sender will also send a small, fixed  
length "index message" encrypted with the recipient's public key.   
The index message will contain the MD5 hash of the full message (and  
a confounder?). The Subject: line for the index message will contain  
an unencrypted copy of the message hash.  The index message will go  
into an index pool.

Instead of downloading the entire message pool to check for messages,  
you download the index pool (should be smaller in size).  You would  
attempt to decrypt each index message (should be quicker).  A  
decryption will yield something that looks like a hash of a full  
message.  Compare this with the contents of the index message's  
Subject: line to detect a successful decryption.

To get the full message, send a request to the message pool server.   
The request will contain a list of message hashes.  One of the hashes  
will be for the message that was sent to you, the other hashes are  
chosen randomly from the collection of index messages you couldn't  
decrypt.

The message pool server will send you the messages that have Subject:  
lines containing the hashes you sent in the request.  You discard all  
the messages that are not for you, decrypt the one that *is* for you,  
and there you have it.

This idea scales up a bit, although not greatly.  I can imagine a  
network of message pools that maintain a distributed index pool among  
themselves.  You can obtain the complete index pool from any of the  
message pool servers.  In this scenario, a index message would  
contain the message hash plus the address of the message pool that is  
holding the associated message.  


Small detail: To avoid downloading sections of the index pool you've  
already seen, the client-side software will need to maintain a  
timestamp or something to keep track of the last index message you've  
seen.  Pass this timestamp to the message pool server to request all  
index messages since "timestamp".

Comments welcome,

Jim_Miller@suite.com