View Issue Details
| ID | Project | Category | View Status | Date Submitted | Last Update |
|---|---|---|---|---|---|
| 0003961 | Composr | core | public | 2019-11-20 17:34 | 2020-01-29 22:28 |
| Reporter | Chris Graham | Assigned To | Chris Graham | ||
| Severity | Feature-request | ||||
| Status | resolved | Resolution | fixed | ||
| Product Version | |||||
| Fixed in Version | |||||
| Summary | 0003961: Probabilistic and Logarithmic update of view counts | ||||
| Description | Composr maintains various view counts (topic view counts, banner view counts, etc). Writing to these each view has a performance impact. It should be a very small impact - but on InnoDB they have to do ACID redo logging in ib_logfile0 and ib_logfile1, multiplying the work needed; and on journaled filesystems, maintenance of the filesystem journal. On a busy site with a stressed disk, this may add up. On MyISAM, there is table-level locking, which has its own performance ramifications. A solution is to reduce our updating of view counts. With a probabilistic model we can reduce updating counts, knowing that they will be statistically similar to the true value if we increase the view tally by the inverse of the likelihood of writing one. To determine the probability of considering a view into the view count we'd do some formula based on the current view count, making sure that the figure is always within some reasonable order of magnitude in terms of accuracy. | ||||
| Tags | Type: Performance | ||||
| Time estimation (hours) | 1.5 | ||||
| Sponsorship open | |||||
|
|
Lines that this will affect (from v9, as this is initially for a customer)... $GLOBALS['SITE_DB']->query('UPDATE '.get_table_prefix().'banners SET views_to=(views_to+1) WHERE '.db_string_equal_to('name',$name),1,NULL,false,true); $GLOBALS['SITE_DB']->query('UPDATE '.get_table_prefix().'banners SET views_from=(views_from+1) WHERE '.db_string_equal_to('name',$name),1,NULL,false,true); $connection->query_update('attachments',array('a_num_downloads'=>$myrow['a_num_downloads']+1,'a_last_downloaded_time'=>time()),array('id'=>$id),'',1,NULL,false,true); $GLOBALS['FORUM_DB']->query('UPDATE '.$GLOBALS['FORUM_DB']->get_table_prefix().'f_topics SET t_num_views=(t_num_views+1) WHERE id='.strval((integer)$this->id),1,NULL,true); $GLOBALS['SITE_DB']->query_update('videos',array('video_views'=>$row['video_views']+1),array('id'=>$row['id']),'',1,NULL,false,true); $GLOBALS['SITE_DB']->query_update('videos',array('video_views'=>$myrow['video_views']),array('id'=>$id),'',1,NULL,false,true); $GLOBALS['SITE_DB']->query_update('images',array('image_views'=>$row['image_views']+1),array('id'=>$row['id']),'',1,NULL,false,true); $GLOBALS['SITE_DB']->query_update('images',array('image_views'=>$myrow['image_views']),array('id'=>$id),'',1,NULL,false,true); $GLOBALS['SITE_DB']->query_update('iotd',array('iotd_views'=>$myrow['iotd_views']),array('id'=>$id),'',1,NULL,false,true); $GLOBALS['SITE_DB']->query_update('catalogue_entries',array('ce_views'=>$entry['ce_views']),array('id'=>$id),'',1,NULL,false,true); $GLOBALS['SITE_DB']->query_update('news',array('news_views'=>$myrow['news_views']),array('id'=>$id),'',1,NULL,false,true); $GLOBALS['SITE_DB']->query_update('galleries',array('gallery_views'=>$myrow['gallery_views']),array('name'=>$cat),'',1,NULL,false,true); $GLOBALS['SITE_DB']->query_update('seedy_pages',array('seedy_views'=>$page['seedy_views']),array('id'=>$id),'',1,NULL,false,true); $GLOBALS['SITE_DB']->query_update('calendar_events',array('e_views'=>$event['e_views']),array('id'=>$id),'',1,NULL,false,true); $GLOBALS['SITE_DB']->query_update('poll',array('poll_views'=>$myrow['poll_views']),array('id'=>$id),'',1,NULL,false,true); if (get_db_type()!='xml') $entry['ce_views']++; $this->connection->query_update('f_members',$change_map+$extra,array('id'=>$id),'',1,NULL,false,true); set_value('page_views',strval(intval(get_value('page_views'))+1)); |
|
|
To stay within an order of magnitude of accuracy (accuracy changing logarithmically), that means a linear relationship between view count and the view count raise step. I plotted some graphs and this works well... Raise step = max(1, view count / 20) Or expressed as probabilities... Probability = min(1, 20 / view count) |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| 2019-11-20 17:34 | Chris Graham | New Issue | |
| 2019-11-20 17:34 | Chris Graham | Tag Attached: Type: Performance | |
| 2019-11-20 21:30 | Guest | Note Added: 0006166 | |
| 2019-11-20 21:45 | Chris Graham | Note Edited: 0006166 | View Revisions |
| 2019-11-20 21:50 | Chris Graham | Note Edited: 0006166 | View Revisions |
| 2019-11-21 02:46 | Chris Graham | Note Added: 0006167 | |
| 2019-11-21 02:47 | Chris Graham | Note Edited: 0006167 | View Revisions |
| 2019-11-21 02:48 | Chris Graham | Description Updated | View Revisions |
| 2019-11-24 02:24 | Chris Graham | Assigned To | => Chris Graham |
| 2019-11-24 02:24 | Chris Graham | Status | non-assigned => resolved |
| 2019-11-24 02:24 | Chris Graham | Resolution | open => fixed |
| 2020-01-29 22:28 | Chris Graham | Description Updated | View Revisions |
