---In [email protected], <fabiano@...> wrote : >>IMHO what is going on is that Firebird is using the lock file (for Classic >>installation) to synchronize between >>various connections (threads in this >>case) to allow one by one to write to FDB file.
I hope that Vlad or Dmitry will tell us how this works for real, so we wouldn't have to guess anymore:) >>I also tested this and if you run various threads inserting among different >>tables there are no performance >>issues. I've tested it briefly and by using two threads and two different tables I can double the performance. However more threads/tables do not seem to speed up things. >>The fastest way I found to insert multiple rows in a Firebird database is run >>a Stored Procedure that >>contains various inserts in form of a long string. >>The SP's splits this string in various inserts. This looks like a variant of method in which all inserts all generated as a big textfile on the client side and then executed on the server like one script. Have you compared your method to this? Because I think your SP_INSERTS might introduce additional overhead (you need build this script on server side). I think both variants might speed things up because you send only one request via Firebird API and network. I will definitely test this out. >>With this >>approach I can insert around 11,000 recods/second! (if there are >>no Indexes, even PK). 11000 records / seconds does not say much, we do not know what system do you have and so on. If you would put this in comparision to the standard method of inserting then we would know how much faster it is. For example, as you see from my test, I am using a standard method and on SSD it is faster than your method: 100000 / 7 seconds gives 14000 records / second. >>I hope I can help you, and sorry about some English mistakes. No need to appologize, I've made many mistakes in my previous post (and in this probably too :) ). Thank you for your help Fabiano! Best regards.
