tag:blogger.com,1999:blog-8673752770146776575.post4001977351484647346..comments2024-01-02T05:04:05.147-08:00Comments on PostgreSQL and Databases in general: Write Scalability in PostgreSQLAmit Kapilahttp://www.blogger.com/profile/01948926447381079550noreply@blogger.comBlogger7125tag:blogger.com,1999:blog-8673752770146776575.post-39461989548761969692015-04-20T22:46:19.485-07:002015-04-20T22:46:19.485-07:00Yeah one part of problem could be in clock sweep a...Yeah one part of problem could be in clock sweep and other part of the problem is that during replacement of page in shared_buffers, we need to flush a page if it is dirty and which is I think true in most write-load cases. We can improve part of it via changing bgwriter settings, but I think still it won't solve the problem. I also think there will be no visible impact in Write Scalability in 9.5. By the way, the tests you have mentioned in your blog post doesn't seem to be pgbench tests which I have ran.Amit Kapilahttps://www.blogger.com/profile/01948926447381079550noreply@blogger.comtag:blogger.com,1999:blog-8673752770146776575.post-1442727219696861732015-04-20T08:25:58.610-07:002015-04-20T08:25:58.610-07:00Well, most likely the problem is inside the clocks...Well, most likely the problem is inside the clocksweep while searching for a victim buffer to replace from shared_buffers. I have described it here [0] and in the bottom there is a stap for seeing that. <br /><br />In that post there also are links to a couple of patches in 9.5 (and you are the reviewer of one of them :)) which should make it a bit better. Your post about read scalability showed that it really makes it better for read-only load (and my testing also confirms it), but it seems that it is not so for heavy writing load :( I haven't finished my testing of 9.5 with writing load yet but early results do not show much difference with 9.4.<br /><br />[0] https://simply.name/postgresql-and-systemtap.htmlvolkhttps://www.blogger.com/profile/11705043540382866904noreply@blogger.comtag:blogger.com,1999:blog-8673752770146776575.post-64241459960311799772015-04-20T07:22:50.934-07:002015-04-20T07:22:50.934-07:00At the default "scale factor" of 1, the ...At the default "scale factor" of 1, the tables initially contain this many rows:<br /><br />table # of rows<br />---------------------------------<br />pgbench_branches 1<br />pgbench_tellers 10<br />pgbench_accounts 100000<br />pgbench_history 0<br /><br />You can (and, for most purposes, probably should) increase the number of rows by using the -s (scale factor) option.<br /><br />For more information on workload and test tool, you can refer PostgreSQL docs:<br />http://www.postgresql.org/docs/devel/static/pgbench.html<br />If you still need more clarification, then do let me know.Amit Kapilahttps://www.blogger.com/profile/01948926447381079550noreply@blogger.comtag:blogger.com,1999:blog-8673752770146776575.post-60426649385709907542015-04-20T05:37:13.579-07:002015-04-20T05:37:13.579-07:00What is the working definition of "scale fact...What is the working definition of "scale factor"?Jamey Hansonnoreply@blogger.comtag:blogger.com,1999:blog-8673752770146776575.post-24896300723655511592015-04-18T08:40:40.144-07:002015-04-18T08:40:40.144-07:00Now it's OKNow it's OK Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-8673752770146776575.post-47556215435561796012015-04-18T02:47:54.973-07:002015-04-18T02:47:54.973-07:00It was visible to me when I checked in safari (Mac...It was visible to me when I checked in safari (Mac browser), but seems there is some problem while opening with other browsers. I have inserted the images, could you verify once again?Amit Kapilahttps://www.blogger.com/profile/01948926447381079550noreply@blogger.comtag:blogger.com,1999:blog-8673752770146776575.post-91738230608357158112015-04-18T02:06:48.430-07:002015-04-18T02:06:48.430-07:00There is something wrong with the link to the imag...There is something wrong with the link to the images, can't see the images.Anonymousnoreply@blogger.com