[image: image.png]
Viet,
For the results of last week's test unfortunately the optimised code is not
statistically significantly different from the non-optimised.
When you send the new test results I can run it again (should take only a
few minutes now I have it set up) and re-run the test. I can also explain
the results if you're interested.
Briefly the test is asking whether the before and after numbers could have
come from the same code (ie no changes). The results are saying that yes,
they could. Statistically the difference in results could (very) easily
have come from the same code. There is no significant difference. In other
words, you could have used only the before code, ran the test twice, and
the difference in results could easily be greater than the difference you
have observed in your test.
Thanks
Joe
On Fri, 10 Sept 2021 at 19:57, Joe Fagan <joe.fa...@bitnine.net> wrote:
> Great Josh.
> I learnt loads reading your script
>
> On Fri, 10 Sep 2021 at 18:14, Josh Innis <josh.in...@bitnine.net> wrote:
>
>> Good Job Viet. I made some modifications to the SQL function that you
>> made. This may be a way to analyze how long it takes to run the queries in
>> a purely postgres way that only calls explain analyze once.
>>
>> SET search_path TO ag_catalog;
>> LOAD 'age';
>> SELECT create_graph('test_graph');
>>
>> CREATE OR REPLACE FUNCTION f_test(ct int, sql text)
>> RETURNS VOID
>> LANGUAGE plpgsql AS $func$
>> DECLARE i int;
>> BEGIN
>> FOR i IN 1 .. $1 LOOP
>> EXECUTE sql;
>> END LOOP;
>> END $func$;
>>
>> EXPLAIN ANALYZE SELECT *
>> FROM f_test( 10000, $q$
>> SELECT * from ag_catalog.cypher('test_graph', $$
>> CREATE (a:Person {id: 129, name: 'Brian', title: 'Researcher',
>> city: 'Munich'})
>> RETURN a
>> $$) as (a agtype)
>> $q$);
>>
>> We create value for our clients by connecting the world's data.
>> Josh Innis Software Engineer
>> Core R&D Team
>> *P*: (831)278-0327
>> 3945 Freedom Circle #260, Santa Clara, CA 95054
>> <https://www.google.com/maps/place/3945+Freedom+Cir+%23260,+Santa+Clara,+CA+95054/>
>> bitnine.net
>>
>>
>> On Fri, Sep 10, 2021 at 9:12 AM John Gemignani <
>> john.gemign...@bitnine.net> wrote:
>>
>>> Good job!
>>>
>>> On Thu, Sep 9, 2021 at 9:11 PM VUONG QUOC Viet <vqv...@bitnine.net>
>>> wrote:
>>>
>>>> Hi Joe and Muhammad,
>>>>
>>>> I've finished the benchmarking experiments after listening to
>>>> Muhammad's suggestions to improve the process: 1. to create 2 separate
>>>> repos for original code and optimized code and 2. place the create_graph at
>>>> the beginning and drop_graph at the end of each sql file.
>>>>
>>>> The experiments were conducted basically the same way: I calculated the
>>>> total time for running 100, 1000 and 10000 sql queries, repeated 20 times
>>>> consecutively, got the average value, and repeated that process 10 times
>>>> more. So in essence, there are totally 200 measurements conducted for
>>>> original and optimized code, each. And I calculated overall average for
>>>> total 200 measurements to get a final number to compare between the
>>>> original and the optimized: So here's the result:
>>>>
>>>> Original code (ms) Optimized code (ms) Difference(ms)
>>>> 100 queries 29.23397 28.786665 0.447305
>>>> 1000 queries 255.391305 250.43654 4.95476500000004
>>>> 10000 queries 2558.612315 2517.9374 40.6749150000001
>>>>
>>>> According to the result, the difference in execution time for 100
>>>> queries is that the optimized code was about 0.45ms faster than the
>>>> original code. For 1000 queries, the optimized code is 4.95ms faster and
>>>> for 10000 queries, it's 40.67ms faster. These numbers, I think, are
>>>> reasonable, because as we make the number of queries 10 times larger, the
>>>> difference in time also increases around 10 times.
>>>>
>>>> I've attached the detailed statistics report. Please let me know your
>>>> comments and suggestions.
>>>>
>>>> Best regards,
>>>> Viet.
>>>>
>>>> --
>
> <http://bitnine.net>
>
> Joe Fagan
>
> Community Advocate Director
>
>
>
> M : +44 7788 148772
>
>
>
> <https://www.facebook.com/search/top?q=bitnine%20agens>
> <http://linkedin.com/company/bitnineglobal> <http://bitnine.net>
> <http://age.apache.org>
> A Graph Extension for PostgreSQL <http://age.apache.org>In the news: Bitnine
> Promoting a First IPO in the Graph Database <https://bwnews.pr/3ds2mUc>
>
