Re: substring on bit(n) and bytea types is slow

On 2 March 2016 at 00:33, Arjen Nienhuis <a(dot)g(dot)nienhuis(at)gmail(dot)com> wrote:

>
> On Feb 29, 2016 22:26, "Evgeny Morozov" <
> evgeny(dot)morozov+list+pgsql(at)shift-technology(dot)com> wrote
> > SELECT substring(bitarray from (32 * (n - 1) + 1) for 32) -- bitarray is
> a column of type bit(64000000)
> > FROM array_test_bit
> > JOIN generate_series(1, 10000) n ON true;
>
> Substring on a bit string is not optimized for long TOASTed values.
> Substring on text is optimized for that. The current code fetches the whole
> 8MB from the table every time.
>
I see, thanks. Is there a better way to pack a large number of integers
efficiently with reasonable read/write performance?

I tried arrays bit varying, which seemed perfect, but in practice when I
stored 4M integers in it, each one taking as few bits as possible, the
table takes 13MB - same as if I just store all of them as bit(24). In fact,
an array of 4M bit(10) integers also takes 13MB. bit(8) takes only 0.7 MB.
bit(9) is where things get weird: for integer 1 to 4M it takes 13MB, but if
I multiple them by 2 (i.e. store 4M even integers) it takes 0.7MB! So there
must be some kind of compression going on there, but I don't understand how
it works.