Add SQL output for some of the test datasets

[cosmoharrigan]

No description provided.

[linas]

I assume these hold data this is equivalent to the other files? Out of curiosity, how quickly do these load?

[cosmoharrigan]

Yes. The load times are:
testscene3.sql <1 second for 3213 atoms
conceptnet4.sql 12 seconds for 139972 atoms, whereas conceptnet4.scm (Scheme) took 354 seconds

[linas]

Ewww. :-( Thanks for posting the data, but it's pretty disappointing.

You seem to be seeing about 12K atoms/sec for the postgres load, and 400 atoms/sec on the scheme load. That's awful. I was measuring .. well, actually I never measured load of text-sql before, so not sure. From sql to atomspace, I was measuring 65K atoms/sec 4 years ago. And according to the benchmark, scheme was last seen doing some 5K-7K atoms-per second, for a complete round-trip (viz enter scheme interpreter, parse the ascii string, do what is says, create the atoms, and return to user.) So maybe the benchmark load is very non-representative of typical data.

I guess we need a benchmark to create 'typical' (random) evaluation links.

[edajade]

It's probably the atomspace event loop. I remember Joel benchmarking that
years ago and saying it slowed things down a lot...

On Mon, Jan 6, 2014 at 10:57 AM, Linas Vepstas notifications@github.comwrote:

Ewww. :-( Thanks for posting the data, but it's pretty disappointing.

You seem to be seeing about 12K atoms/sec for the postgres load, and 400
atoms/sec on the scheme load. That's awful. I was measuring .. well,
actually I never measured load of text-sql before, so not sure. From sql to
atomspace, I was measuring 65K atoms/sec 4 years ago. And according to the
benchmark, scheme was last seen doing some 5K-7K atoms-per second, for a
complete round-trip (viz enter scheme interpreter, parse the ascii string,
do what is says, create the atoms, and return to user.) So maybe the
benchmark load is very non-representative of typical data.

I guess we need a benchmark to create 'typical' (random) evaluation links.

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/7#issuecomment-31619868
.

[linas]

On 7 January 2014 02:46, jadeoneill notifications@github.com wrote:

It's probably the atomspace event loop. I remember Joel benchmarking that
years ago and saying it slowed things down a lot...

That atomspace event loop is no more :-) and yes, removing it speeded
things up by 5x or more. Precise numbers in the diary file in the
benchmarking directory.

--linas