lua-users home
lua-l archive

[Date Prev][Date Next][Thread Prev][Thread Next] [Date Index] [Thread Index]


On Wed, Oct 2, 2013 at 2:27 PM, Michal Kottman <k0mpjut0r@gmail.com> wrote:
> If anyone is interested in more precise measurements (or other key-value
> databases), please point out how I can improve the methodology, or provide
> some real-world use cases.


one thing bothers me:

obviously, a Lua table should be the fastest, both because it works
directly in the VM and also because it works purely in RAM.  the
surprising thing is that the other options are in the same ballpark of
performance.

but it shouldn't be so surprising, since if the benchmark is the same
size for all databases, that means that the whole data does fit in
RAM, invalidating the test!

i propose two modifications:

1: make sure that the databases are _MUCH_ bigger, at least 3-6x times
the available RAM

2: don't do the read benchmark immediately after the write.  at least
close and open the DB, even better if it's on a separate run.

if after that any DB manages to hit more than 1/10 the OP/s of Lua
tables, I'll be very surprised

-- 
Javier