tag:blogger.com,1999:blog-12493181.post2338367185965955517..comments2022-03-30T21:05:05.842+09:00Comments on A Hacker's Diary: pgbench on UNLOGGED table(s), Round 2Satoshi Nagayasuhttp://www.blogger.com/profile/09706093898823214408noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-12493181.post-35330644460599762352011-10-08T10:48:48.677+09:002011-10-08T10:48:48.677+09:00Indeed. Thanks for pointing out about that.
The d...Indeed. Thanks for pointing out about that.<br /><br />The difference may be occurred by not only write i/o itself, but also lock stuffs in parallel processing. Lock contention is never happen with using unlogged tables.<br /><br />Anyway, I think unlogged and async are never happen at the same time if async means wal i/o, because there're no wal i/o with "unlogged". I understand Satoshi Nagayasuhttps://www.blogger.com/profile/09706093898823214408noreply@blogger.comtag:blogger.com,1999:blog-12493181.post-18086251394232923172011-10-07T23:18:36.891+09:002011-10-07T23:18:36.891+09:00I think the difference may show up more when havin...I think the difference may show up more when having a lot of parallel activity going on. Not writing wal with unlogged should create somewhat less disk i/o than just delaying the i/o with async commit or not forcing commits.<br /><br />And then unlogged + async may be most interesting for tables with lots of writes of data that could be lost on crash?Heikohttps://www.blogger.com/profile/04989356067273734299noreply@blogger.com