Re: [Caml-list] [ANN] Datalog-0.1

[Simon Cruanes <simon.cruanes.2007@m4x.org>]

So, I benchmarked it against the Lua datalog engine
(http://www.ccs.neu.edu/home/ramsdell/tools/datalog/) on two kinds of
problems (that can already be found in the tests/ directory, with
scripts to generate problems of any size):

- graph: a transitive reflexive relation on a graph, 'reachable()', an a
relation 'edge()' that describes a graph of size n. The graph is a big
cycle (ie E_i -> E_{i+1}, and E_n -> E_1). The transitive closure is
therefore a clique of size n. The query is 'reachable(X,Y)?', which
returns n^2 facts.

- induction (bad name): two mutually recursive predicates, p(i) :- q(i),
p(i-1) and q(i) :- p(i-1), q(i-1) for i from 1 to n, and initial facts
p(0) and q(0). The query is 'P(X)?', and returns n facts.

The respective command lines used to benchmark are as follow
('datalog_cli' is in OCaml, 'datalog' is the lua engine):

$ time datalog_cli tests/graph500.pl -pattern 'reachable(X,Y)' > /dev/null
$ time echo 'reachable(X,Y)?' | datalog tests/graph500.pl -i > /dev/null

$ time datalog_cli tests/induction100.pl -pattern 'p(X)' > /dev/null
$ time echo 'p(X)?' | datalog tests/induction100.pl -i > /dev/null

I could not find an easy way to measure the computation time, without
printing the solution (which is also heavy); however, in both cases the
full solution is printed.

So, here are the respective times (total time) on a Intel i5 3.2GHz,
with 4GB of RAM:

problem          datalog_cli  datalog
--------------------------------------------
graph200.pl      0.289s       0,798s
graph500.pl      3.005s       5.092
graph1000.pl     19.151s      21.138s
graph1500.pl     52.509s      51.589s
graph2000.pl     151.04s      OOM (swapped, became too slow)
induction200.pl  0.024s       0.138s
induction500.pl  0.021s       0.750s
induction1000.pl 0.066s       2.997s
induction1500.pl 0.075s       6.911s
induction2000.pl 0.134s       12.856s

So, the OCaml library looks competitive with the Lua SLD-resolution
engine. I believe it performs better on the 'induction' examples because
of the memorization of intermediate results. On the other hand, it
always compute the whole fixpoint, whereas SLD-resolution can focus on
what is needed for the query.

Hope that helps! :)

Cheers,

--
Simon Cruanes

On 01/28/2013 07:06 PM, yoann padioleau wrote:
> Hi,
> 
> How does it compare in terms of performance with other datalog engines?
> 
> On Mon, Jan 28, 2013 at 3:11 AM, Simon Cruanes
> <simon.cruanes.2007@m4x.org> wrote:
>> Hello,
>>
>> I'm pleased to announce the first (alpha) release of Datalog, a small
>> fixpoint engine for the Datalog fragment of logic
>> (http://en.wikipedia.org/wiki/Datalog) written in OCaml. The library is
>> designed to be used to compute fixpoints of rules, incrementally. It is
>> available under the BSD license at https://github.com/c-cube/datalog .
>>
>> Both a command-line fixpoint computation tool and a functorial library
>> are provided. Input files have a prolog-like syntax, but without
>> function symbols. The command line tool works like this (the example
>> computes the transitive closure of a graph, and then cliques within this
>> transitive closure):
>>
>> $ cat tests/clique10.pl
>> reachable(X,Y) :- edge(X,Y).
>> reachable(X,Y) :- edge(X,Z), reachable(Z,Y).
>> same_clique(X,Y) :- reachable(X,Y), reachable(Y,X).
>> edge(0, 1).  % edge from 0 to 1
>> edge(1, 2).
>> edge(2, 3).
>> edge(3, 4).
>> edge(4, 5).
>> edge(5, 0).
>> edge(5, 6).
>> edge(6, 7).
>> edge(7, 8).
>> edge(8, 9).
>> edge(9, 10).
>> edge(10, 7).
>>
>> $ ./datalog_cli.native tests/clique10.pl -pattern 'same_clique(1,X)'
>> % start datalog
>> % parse file tests/clique10.pl
>> % process 15 rules
>> % computing fixpoint...
>> % done.
>> % facts matching pattern same_clique(1, X1):
>> same_clique(1, 0).
>> same_clique(1, 1).
>> same_clique(1, 3).
>> same_clique(1, 2).
>> same_clique(1, 5).
>> same_clique(1, 4).
>> % max_heap_size: 126976; minor_collections: 0; major collections: 0
>>
>> The library provides a functor that allows one to create rules (Datalog
>> clauses), and add them to a structure that computes their fixpoint
>> incrementally (i.e., the fixpoint is updated after each rule is
>> inserted). Callbacks can be associated to symbols, to be called whenever
>> a fact is derived.
>>
>> Feedback, comments, or bug reports are very welcome!
>>
>> Cheers,
>>
>> --
>> Simon Cruanes
>>
>>
>> --
>> Caml-list mailing list.  Subscription management and archives:
>> https://sympa.inria.fr/sympa/arc/caml-list
>> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
>> Bug reports: http://caml.inria.fr/bin/caml-bugs
>