Slimelite BBS

what is a serial number?

Back to FREE25

From Pierre Phaneuf · 1996-04-11 07:09 · #73152 in FREE25

In reply to what is a serial number? by HECTOR SANTOS

 HS> Well, first, there is no guarantee this "counter" will be unique
 HS> across other systems.  But lets just concern ourselves with what
 HS> FTS-9 says "Unique on the original system" only.  That is what it
 HS> says. Right?

Why bother with the uniqueness on other systems? There is the site address right beside the serial number and it's there for a reason! If you only compare the serial numbers of messages when checking for dupes, you are doing it wrong just for this reason... The MSGID string in whole (the address and the serial number) is a single string to be directly compared... I guess you could do it in two parts (check out only the serial number and when two matches, check out the addresses) to do it faster, but leaving the address out is wrong... It is also why including the origin address in your CRC calculation is not worth much, it's duplicating information...

 HS> This presents a multiple of problems since in order to guarantee
 HS> uniqueness, all email generating processes in the system whether its
 HS> coming from a multi-node, multi-threaded system,  3rd party door
 HS> system, they all must have a "COMMON SERVER", if you will, to retrieve
 HS> the last number used.  Do you see the picture? Do you understand this
 HS> problem?

Yes, this is the hard part, I agree with you, but I'm perplexed by this part...

 HS> BTW, new Client/Server software do allow for this to happen now.
 HS> Wildcat 5.0 requires that you use the SERVER to create or toss a
 HS> message and it will always return a unique message id for its own dupe
 HS> checking system. This is all relative to the installation. There is no
 HS> way to retoss the same message.  No way. Obviously, the vendor has
 HS> factored in some "parameters" of the message into the "equation".

You're saying that the main problem with serial number is that they should come from a single server and if it was not for that, it would be good, then you are telling me that WildCat 5.0 has such a unique server and that it should not use serial numbers??? If I'd design such a client/server type BBS, I'd jump on the occasion to just put the serial number counter in the server and let it roll that way...

 HS> In my opinion, this is ideal and is required. By using a serial
 HS> number counter, you have effectively removed or excluded the
 HS> parameters of the message itself and this increases the likelyhood
 HS> that two SAME messages can be theoritically tossed into the system.

Look Hector, I know this is highly simplified, but here is why CRC and "including parameters of the message" is bad: consider this equation would be our serial number generator: x+y=serial number, x and y being parameters of the message. If we have 1+4, it gives 5. If we have 2+1, it gives 3. Not bad, but if we have 2+3, it gives 5 again even if the two parameters are totally different. CRC is more resistant to CRC-neutral difference than a checksum, but nevertheless, CRC-neutral strings *exist* and can happen at any moment, be it today or in twelve years.

 HS> Well, unless two systems USE the same "implementation" there is no
 HS> way that one can guarantee that another implementation will NOT
 HS> "REPRODUCE" a serial number that was already another system and sent
 HS> to your system.

What would I care that another system reproduce one (or all if they like!) of my serial number! When checking for dupes, you must compare the MSGIDs. A MSGID is composed of a serial number, but also of the site address. If site A generate a message #1 with the same serial number that site B gave to message #2, I will not tag them as dupes, since they don't have the same origin address (obviously, dupes generally comes at least from the same machine!)...

 HS> So my point is, you can go directly to the specs and attempt to
 HS> follow it and you will see that many things, there are issues that
 HS> doesn't quick make sense.  Remember, it is not the bible.  The
 HS> authors of the specs make great contributions but they are also not
 HS> great sages with perfect vision into the future.  Many times, it just
 HS> doesn't quite work.  If it did, we would be having many of the
 HS> debates in this echo.

I see that if a produce a serial number that is unique on my system for at least a significant length of time and that I have a site address that is unique in the network where I am (I hope so for mail delivery!), dupes potentially produced by my system will be detected and taken care of by dupe checker looking for the MSGID, for as long as my serial number doesn't wrap. And if you take the bible as making sense, I understand you better... :-)

 HS> In this echo, we get LOST of "serial numbers" and the chaining of the
 HS> FTS-9 and it has hardly any impact in the fido market. Almost none
 HS> whatsoever. We throw away the important of addressing.

For once, I think I got an analogy that's not rotten... Let's say the DMV would put your complete phone number, with 2 or 3 additional letters, to put on license plates. Some of us could think and tell themselves "wow, if I need to contact Bob, I'll take a look at his license plate, note it, remove the extra letters and call him!"... That would be good most of the time... Except that in fact, the license plate identifies the car, not the driver, so when if Bob buyed his car second-hand, it's not his phone number on the license plate, but the one of the original owner! (who would not be very happy of those thinking about stripping some letters of something not meant to contact the owner)

Pierre
phaneuf@absint.login.qc.ca

--- FMail/386 1.02
 * Origin: Zaphod Beeblebrox for President! (1:167/136.14)