[NNTP] 64-bit article counter extension strawman
Steve Walker
nntp at nntpserver.com
Tue Jul 19 08:14:26 PDT 2005
Charles Lindsey wrote:
>In <42DBEF3A.7050305 at nntpserver.com> Steve Walker <nntp at nntpserver.com> writes:
>
>
>
>>Russ Allbery wrote:
>>
>>
>
>
>
>>>Do we want to allow rolling over? Rolling over comes with its own set of
>>>bugs, in particular triggering the low > high logic that causes many
>>>
>>>
>
>
>
>>I think rolling over is prone to problems and it's not worth it
>>when there is there is a good clean method of using 64 bit numbers.
>>
>>
>
>And I think exactly the opposite. It is ridiculous using huge numbers
>(approaching 64 bits) to identify articles in a group where the maximum
>number of articles actually existing at any one time is unlikely to exceed
>100,000, and probably nowhere near that.
>
>
A few years ago and I would have agreed. Today, it's not unreasonable
to have a binary group with 100 million active articles. If your server
software assigns unique article numbers you also will use numbers much
faster as all articles are in a single "master" group. This server
design has some very good benefits with the only downside being that you
will reach the 32 bit limit in about 2 years on a full feed server.
>So we should look at wrap-around solutions (several variants have been
>mentioned). What are the problems? Special meanings for zero, and several
>ways of detecting whether a group is empty. Is that it?
>
>
Clients have to add a lot of logic for detecting the wraparound. The
number zero is a reserved article number, but to do simple % math we
really need it as a valid number. The xover command has to support a
high < low number request. In general, it will be prone to errors.
Wrapping can work, but it would be complex.
Steve.
More information about the ietf-nntp
mailing list