Re: Maximum table size

On Tue, 9 Sep 2003 14:25:19 -0400 (EDT), pgman(at)candle(dot)pha(dot)pa(dot)us (Bruce
Momjian) wrote:

>Tatsuo Ishii wrote:
>> > Tom Lane wrote:
>> > > Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:
>> > > > Is our maximum table size limited by the maximum block number?
>> > >
>> > > Certainly.
>> > >
>> > > > Is the 16TB number a hold-over from when we weren't sure block number
>> > > > was unsigned, though now we are pretty sure it is handled as unsigned
>> > > > consistenly?
>> > >
>> > > It's a holdover. As to how certain we are that all the
>> > > signed-vs-unsigned bugs are fixed, who have you heard from running a
>> > > greater-than-16Tb table? And how often have they done CLUSTER, REINDEX,
>> > > or even VACUUM FULL on it? AFAIK we have zero field experience to
>> > > justify promising that it works.
>> > >
>> > > We can surely fix any such bugs that get reported, but we haven't got
>> > > any infrastructure that would find or prevent 'em.
>> >
>> > I guess the big question is what do we report as the maximum table size?
>> > Do we report 32TB and fix any bug that happen over 16TB?
>>
>> That seems right direction for me. I see no reason why 16TB is more
>> reliable number than 32TB, since nobody has ever tried to build 16TB
>> tables.
>
>Agreed. I think the question is how large does the design support,
>rather than how large have we tested. (In fact, the check for using
>block numbers as unsigned was removed from the FAQ when I reviewed the
>code.)
>
>I know Tom is concerned because we haven't tested it, but I don't think
>anyone has tested 16TB either, nor our 1600-column limit.

Well, made some tests with 1600 shall not be so difficult and I'll not
bet that nobody reached this limit

>
>Also, I think people look at these numbers to determine if PostgreSQL
>can handle their data needs 5-10 years down the road.

I don't agree that people are looking at PostgreSQL fot handle 5-10
years old, what I think ( is anyway my opinion ) is that people are
looking at postgres in order to avoid more expensive tools like
ORACLE, SYBASE, INFORMIX, and have a low TCO

>In fact, if you increase the page size, you can quadruple most of the
>existing limits. This is already mentioned in the FAQ:
>
> <P>The maximum table size and maximum number of columns can
> be increased if the default block size is increased to 32k.</P>
>
>I have updated the FAQ to say 32TB, and of course, larger page sizes
>could make this 128TB.

Why this ? just because bigger is better? I agree with Tom Lane, is
better underpromise than overpromise.

Regards
Gaetano Mendola