Wikipedia statistics benchmark

There several goals of benchmark

• Compare engines in OLAP queries for planning, predicting growth, analyzing access patterns to wiki pages, draw trends.
• Compare engines in statistical queries for end users, which can be executed in real-time. I.e. How many times that or another page was accessed yesterday vs today.
• Understand specific features and characteristic of each engine.
• Compare throughput on simple queries
• Check ability to load data and serve queries at the same time ( availability during data load )

It's important for benchmark to use real data. I am going to take statistics of access to WikiPedia pages http://lists.wikimedia.org/pipermail/wikitech-l/2007-December/035435.html

Format of raw data is

PROJECT PAGE ACCESSED BYTES
...
aa Category:User_aa 1 6313                                                                                               
aa Main_Page 3 16473                                                                                                     
aa Special:RecentChangesLinked/User:Purodha 1 5701                                                                       
aa Special:Statistics 1 661                                                                                              
aa Talk:Main_Page 1 8092  
...

I made some efforts to normalize data, final tables you can see below

Normalized data is available in AWS snapshot (“percona-wikistat”, ID:snap-a5f9bacc)

• pages

Dictionary of pages. 724.550.811 rows. data size: 40476M

CREATE TABLE `pages` (
  `id` int(10) unsigned NOT NULL AUTO_INCREMENT,
  `page` varchar(1024) NOT NULL,
  PRIMARY KEY (`id`)
)

• projects

Projects dictionary, small table. 2025 rows.

CREATE TABLE `projects` (
  `id` smallint unsigned NOT NULL AUTO_INCREMENT,
  `project` varchar(50) NOT NULL,
  PRIMARY KEY (`id`)
)

• datesinfo

Table with different part of date. 9624 rows.

CREATE TABLE `datesinfo` (
  `id` smallint unsigned NOT NULL AUTO_INCREMENT,
  `datestring` char(11) NOT NULL,
  `caldate` date NOT NULL,
  `calday` int(10) unsigned NOT NULL,
  `calmonth` tinyint(3) unsigned NOT NULL,
  `calyear` smallint(5) unsigned NOT NULL,
  `dayofweek` tinyint(3) unsigned NOT NULL,
  `dayhour` tinyint(3) unsigned NOT NULL,
  PRIMARY KEY (`id`)
)
+------+-------------+------------+--------+----------+---------+-----------+---------+
| id   | datestring  | caldate    | calday | calmonth | calyear | dayofweek | dayhour |
+------+-------------+------------+--------+----------+---------+-----------+---------+
...
| 3625 | 20090601-00 | 2009-06-01 |      1 |        6 |    2009 |         2 |       0 | 
| 3626 | 20090601-01 | 2009-06-01 |      1 |        6 |    2009 |         2 |       1 | 
| 3627 | 20090601-02 | 2009-06-01 |      1 |        6 |    2009 |         2 |       2 | 
| 3628 | 20090601-03 | 2009-06-01 |      1 |        6 |    2009 |         2 |       3 | 
| 3629 | 20090601-04 | 2009-06-01 |      1 |        6 |    2009 |         2 |       4 | 
| 3630 | 20090601-05 | 2009-06-01 |      1 |        6 |    2009 |         2 |       5 | 
| 3631 | 20090601-06 | 2009-06-01 |      1 |        6 |    2009 |         2 |       6 | 
| 3632 | 20090601-07 | 2009-06-01 |      1 |        6 |    2009 |         2 |       7 | 
| 3633 | 20090601-08 | 2009-06-01 |      1 |        6 |    2009 |         2 |       8 | 
| 3634 | 20090601-09 | 2009-06-01 |      1 |        6 |    2009 |         2 |       9 | 
...
+------+-------------+------------+--------+----------+---------+-----------+---------+

• pagestat

Fact table. page_count is aggregated amount (sum) of page calls per hour.

Raw files per month:

1. 200906: 68.352M / 3.453.013.109 rows
2. 200907: 68.152M / 3.442.375.618 rows
3. 200908: 69063M
4. 200909: 69480M
5. 200910: 77.558M / 3.919.634.163 rows
6. 200911: 74.851M /
7. 200912: TBD

CREATE TABLE `pagestat` (
  `date_id` smallint unsigned NOT NULL,
  `project_id` smallint unsigned NOT NULL,
  `page_id` int unsigned NOT NULL,
  `page_count` mediumint unsigned NOT NULL,
)

Queries under consideration:

• Q_UNIQ(Year, Month):

  
   SELECT count(distinct page_id) FROM pagestat JOIN datesinfo di ON ( di.id=date_id ) WHERE di.calmonth=6 and di.calyear=2009;

• Q_DAYSTAT(Year, Month):

   SELECT caldate, sum(page_count) 
   FROM pagestat JOIN datesinfo di ON ( di.id=date_id ) 
   WHERE di.calmonth=7 and di.calyear=2009 
   GROUP BY caldate ORDER BY caldate;

• Q_TOP_PROJECTS(N,Year,Month):

   SELECT project, sum(page_count) sm 
   FROM pagestat 
     JOIN datesinfo di ON ( di.id=date_id ) 
     JOIN projects p ON  (p.id=project_id ) 
   WHERE di.calmonth=7 and di.calyear=2009 
   GROUP BY project 
   ORDER BY sm DESC 
   LIMIT 20;
   

• Q_PROJECT_SUM

   SELECT project_id, sum(page_count) sm
   FROM pagestat
   GROUP BY project_id;

• Q_HOURPROJECTSTAT(project,Year,Month,Day):

   SELECT dayhour, sum(page_count) sm 
   FROM pagestat 
     JOIN datesinfo di ON ( di.id=date_id ) 
     JOIN projects p ON  (p.id=project_id ) 
   WHERE di.caldate='2009-07-21' AND project='es' 
   GROUP BY dayhour;

* Q_NOPROJECTS(Year,Month,Day)

   SELECT project FROM projects LEFT OUTER JOIN pagestat ON (pagestat.project_id = projects.id) WHERE pagestat.project_id IS NULL ORDER BY project; 
   

• Q_NOPROJECTS_1(Year,Month,Day):

   SELECT project FROM projects 
    LEFT OUTER JOIN 
     (SELECT distinct project_id pid FROM pagestat WHERE date_id IN 
      (SELECT id FROM datesinfo WHERE calmonth=7 and calyear=2009 AND calday=14  )) t1 
    ON (projects.id=pid) 
   WHERE pid IS NULL ORDER BY project;

• Q_DAYWEEKSTAT(project,Year,Month)

   SELECT dayofweek, sum(page_count) sm 
   FROM pagestat 
     JOIN datesinfo di ON ( di.id=date_id ) 
     JOIN projects p ON  (p.id=project_id ) 
   WHERE di.calmonth=6 and di.calyear=2009 AND project='de' 
   GROUP BY dayofweek;  

• Q_TOP50(project,Year,Month,Day)

   SELECT page, sum(page_count) sm 
   FROM pagestat 
     JOIN datesinfo di ON ( di.id=date_id ) 
     JOIN projects p ON  (p.id=project_id ) 
     JOIN pages pp ON (pp.id=page_id) 
   WHERE di.calmonth=6 and di.calyear=2009 AND di.calday=18 
     AND project='de' 
   GROUP BY page 
   ORDER BY sm DESC 
   LIMIT 50;

• Q_MAXPAGE(Year,Month,Day)

   SELECT page_id, max(page_count) mx
   FROM pagestat
     JOIN datesinfo di ON ( di.id=date_id ) 
   WHERE di.calmonth=6 and di.calyear=2009 AND di.calday=18 
   GROUP BY page_id
   ORDER BY mx DESC
   LIMIT 50;

• Q_10000HITS_1(Year,Month,Day) - found pages with more than 10000 hits per hour

   SELECT page_id, max(page_count) mx
   FROM pagestat
     JOIN datesinfo di ON ( di.id=date_id ) 
   WHERE di.calmonth=7 and di.calyear=2009 AND di.calday=18 
   GROUP BY page_id
   HAVING max(page_count) > 10000
   LIMIT 50;
  

• Q_10000HITS_2(Year,Month,Day) - found pages with more than 10000 hits per hour

   SELECT page,mx FROM pages 
    JOIN (SELECT page_id, max(page_count) mx  
          FROM pagestat    
          JOIN datesinfo di ON ( di.id=date_id )   
          WHERE di.calmonth=7 and di.calyear=2009 AND di.calday=18   
          GROUP BY page_id  
          HAVING max(page_count) > 10000  LIMIT 50) t 
    ON (t.page_id=pages.id);
 

• Q_NOACCESS_SECOND_DAY(project_id, Year, Month, Dday): Find pages accessed more 1000 times one day and not accessed another

   SELECT page FROM pages JOIN 
    (SELECT t1.page_id FROM 
      (SELECT p1.page_id,sum(page_count) sm FROM pagestat p1 JOIN datesinfo di1 ON ( di1.id=p1.date_id ) 
       WHERE di1.calmonth=6 and di1.calyear=2009 AND di1.calday=18 AND p1.project_id=193 
       GROUP BY p1.page_id 
       HAVING sum(page_count)>1000)  t1 
      LEFT OUTER JOIN  
      (SELECT p2.page_id FROM pagestat p2 JOIN datesinfo di2 ON ( di2.id=p2.date_id ) 
       WHERE di2.calmonth=6 and di2.calyear=2009 AND di2.calday=19 AND p2.project_id=193) t2 
      ON (t2.page_id=t1.page_id)  WHERE t2.page_id IS NULL) bq 
   ON (pages.id=page_id);

• Q_PAGE_MONTHSTATS(page,Year, Month)

   SELECT caldate, sum(page_count) FROM pagestat JOIN datesinfo ON ( pagestat.date_id = datesinfo.id)
   JOIN pages ON (pagestat.page_id=pages.id) WHERE page='Yugopolis' AND calmonth=7 and calyear=2009 GROUP by caldate ORDER BY caldate;

• Q_PROJECT_AVG(Year, Month)

   SELECT project, AVG(page_count) FROM pagestat JOIN datesinfo ON ( pagestat.date_id = datesinfo.id)
   JOIN projects ON (pagestat.project_id=projects.id) WHERE calmonth=7 AND calyear=2009 GROUP by project;

• Q_PAGES_PROJECT(Year,Month) Found pages that belong to several projects

   SELECT page_id,  count(distinct project_id) FROM pagestat JOIN datesinfo di ON ( di.id=date_id ) 
   WHERE di.calmonth=7 and di.calyear=2009 GROUP BY page_id HAVING count(distinct project_id) > 10 LIMIT 100;

• Q_HOURS_DROP(Date, Hour, project_id): Find page that one hour was accessed two times more than next hour

   SELECT ps1.page_id, ps1.page_count 
   FROM pagestat ps1 
    JOIN datesinfo di1 ON ( ps1.date_id = di1.id) 
    JOIN pagestat ps2 ON (ps2.page_id=ps1.page_id) 
    JOIN  datesinfo di2 ON ( ps2.date_id = di2.id) 
  WHERE di1.caldate='2009-07-20' AND di1.dayhour=14 
    AND di2.caldate='2009-07-20' AND di2.dayhour=15 
    AND ps1.project_id=193 AND ps2.project_id=193 
    AND ps1.page_count > 2 * ps2.page_count 
    AND ps1.page_count>=1000 LIMIT 100; 

• Build summary tables with per-day statistics

CREATE TABLE pagestat_daily (
   page_id int NOT NULL,
   date_id_from smallint  NOT NULL,
   date_id_to smallint  NOT NULL,
   project_id smallint  NOT NULL,
   page_count int  NOT NULL );

INSERT INTO pagestat_daily (project_id, page_id, date_id_from, date_id_to, page_count ) 
SELECT project_id, page_id, min(date_id), max(date_id), sum(page_count) 
FROM pagestat 
JOIN datesinfo ON (date_id=id) 
WHERE calmonth=6 AND calyear=2009 GROUP BY project_id, page_id,caldate;

• Q_DATE_RANGE - date range query on summary table

   SELECT project, sum(page_count) sm 
   FROM pagestat_daily 
    JOIN datesinfo dstart ON ( date_id_from=dstart.id) 
    JOIN datesinfo dend ON (date_id_to=dend.id) 
    JOIN projects ON (project_id=projects.id) 
   WHERE dstart.caldate>='2009-06-15' AND dend.caldate <= '2009-06-20' 
   GROUP BY project HAVING sum(page_count) > 1000 
   ORDER BY sm DESC; 

• Q_PAGE_DATE_RANGE - date range on summary table per page

   SELECT page, sm FROM pages, 
   ( SELECT page_id, sum(page_count) sm 
     FROM pagestat_daily 
      JOIN datesinfo dstart ON ( date_id_from=dstart.id) 
      JOIN datesinfo dend ON (date_id_to=dend.id) 
    WHERE dstart.caldate>='2009-06-15' AND dend.caldate <= '2009-06-20' 
   GROUP BY page_id ) dt 
   WHERE pages.id=dt.page_id ORDER BY sm DESC LIMIT 100;

• Q_HIT_SPIKE - show max per hour spikes

   SELECT ps.page_id, MAX(ps.page_count/pd.page_count) AS max_daily_spikiness
   FROM pagestat ps
   INNER JOIN pagestat_daily pd ON ps.page_id=pd.page_id
     AND ps.date_id BETWEEN pd.date_id_from AND pd.date_id_to
   GROUP BY ps.page_id
   ORDER BY max_daily_spikiness DESC
   LIMIT 10;

• QC_POINT_FACT Retrieve specific page on specific date from fact table

   SELECT * FROM pagestat ps1 
   JOIN datesinfo di1 ON ( ps1.date_id = di1.id) 
   WHERE page_id=64193807 AND project_id=193 
   AND di1.calday=20 AND di1.calmonth=7 
   AND di1.calyear=2009 AND di1.dayhour=15;

• QC_POINT_PAGE(page_id)

 SELECT * FROM pages WHERE id=253608709;

* Retrieve stats for page / project for last hour

Engines in consideration:

• InfoBright, ICE 3.3
• InfiniDB, 0.9.5.4-beta
• MonetDB, Nov-2010
• LucidDB
• Greenplum, 3.3.4

CREATE TABLE pages (  id int NOT NULL , page varchar(1024) NOT NULL )
COPY 730000000 records INTO pages FROM '/data/vol2/wikistat/in/pages.txt';

Install pcre-6.6 with UTF-8 support ( http://www.fredsantos.net/index.php?option=com_content&view=article&id=114:unicode-support-on-centos-52-with-php-and-pcre&catid=36:linux&Itemid=77 )

rpm -ivh http://mirrors.kernel.org/centos/5.4/os/SRPMS/pcre-6.6-2.el5_1.7.src.rpm
change to %configure --enable-utf8 --enable-unicode-properties in pcre.spec

or

rpm -ivh http://gaarai.com/wp-content/uploads/2009/01/pcre-66-27x86_64.rpm

/etc/sysctl.conf
kernel.shmmax=500000000
kernel.shmmni=4096
kernel.shmall=4000000000
kernel.sem=250 64000 100 512
net.ipv4.tcp_tw_recycle=1

CREATE TABLE pages (  id int NOT NULL PRIMARY KEY, page varchar(1024) NOT NULL );
CREATE TABLE pagestat (
   date_id smallint  NOT NULL,
   project_id smallint  NOT NULL,
   page_id int NOT NULL,
   page_count int  NOT NULL )
WITH (appendonly=true, orientation=column, compresslevel=1, compresstype=QuickLZ);


gpfdist -d /mnt/data/ -p 8081 -l /home/gpadmin/log

CREATE EXTERNAL TABLE ext_pages ( id int, 
   page varchar(1024) ) 
   LOCATION ('gpfdist://localhost:8081/pages.txt')
 FORMAT 'TEXT' (DELIMITER '|');

CREATE EXTERNAL TABLE ext_pagestat (   date_id smallint , project_id smallint, page_id int ,page_count int )
LOCATION ('gpfdist://localhost:8081/200906.out') 
FORMAT 'TEXT' (DELIMITER '|');