Re: Yet one more stupid question
От | Dmitri Touretsky |
---|---|
Тема | Re: Yet one more stupid question |
Дата | |
Msg-id | 70275739001.20021205091346@listsoft.ru обсуждение исходный текст |
Ответ на | Re: Yet one more stupid question (Ron Johnson <ron.l.johnson@cox.net>) |
Список | pgsql-novice |
Good time of the day! >> RJ> Slowdown when inserting into a high-tuple table vs. a low-tuple >> RJ> table? >> >> Yes. And the difference is pretty big... Or I misunderstood your >> question? RJ> When the table had 1,000 rows in it, inserting was faster than now RJ> that it has 20M records? Sounds like you understood me... RJ> I've done the same thing and didn't notice all that much of a drop- RJ> off in speed. (IDE drive using the EXT3 filesystem.) I use SCSI drive with UFS. Perhaps the slowdown comes from multiple simultaneous INSERTs - I've tried to record server visits into table (something like 70-100 thousand records a day). It works fine for one-two persons (even with loaded table) and works fine under load with "empty" table (up to around 500 - 700 thousand records). After that every insert starts to require like 3-5 seconds, which causes locks for other processes (I have about 100-150 connections) - every INSERT is a transaction and other processes can't add record till this transaction finish. Slowdown is steady, but at some point server just hangs - I belive it happens when time needed for transaction to finish is increased to the point when new visitors can't connect to the database (due to no available connections) and receive "too many clients already". Best regards, Dmitri ( mailto:dmitri@listsoft.ru ) New SOFT daily (RUS): http://www.listsoft.ru/ (ENG): http://www.listsoft.com/ Articles, tips : http://www.diskovod.ru/ --- Computer - A device designed to speed and automate errors.
В списке pgsql-novice по дате отправления: