> Compressor | level | heap size | toastrel | toastidx | seconds
> | | | size | size |
> -----------+-------+-----------+----------+----------+--------
> PGLZ | - | 425,984 | 950,272 | 32,768 | 5.20
> zlib | 1 | 499,712 | 614,400 | 16,384 | 6.85
> zlib | 3 | 499,712 | 557,056 | 16,384 | 6.75
> zlib | 6 | 491,520 | 524,288 | 16,384 | 7.10
> zlib | 9 | 491,520 | 524,288 | 16,384 | 7.21
Consider that the 25% slowness gets us a 35% disk reduction, and that
translates to fewer buffer blocks and disk accesses. Seems there is a
clear tradeoff there.
> If replacing the compressor/decompressor can cause a runtime
> difference of 25% in such a scenario, the pure difference
> between the two methods must be alot.
>
> Leave PGLZ in place as the default compressor for toastable
> types. Speed is what all benchmarks talk about - on disk
> storage size is seldom a minor note.
True, disk storage is not the issue, but disk access are an issue.
> We can discuss about enabling zlib as a per attribute
> configurable alternative further. But is the confusion this
> might cause worth it all?
I think we have to choose one proposal.
-- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610)
853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill,
Pennsylvania19026
