Re: Optimal buffer flushing

>From: pef@alert.com (Peter E. Farley) >Subject: Optimal buffer flushing >Date: Fri, 21 Oct 1994 23:48:38 GMT >X-Informix-List-Id: <news.9390> >Question: I can find no documentation to help me. What is the optimal >number of rows to place in the insert buffer before flushing? With my >current program a flush is done every thousand regardless of the number of >rows in the insert buffer. The buffer used is of finite size (non-configurable), and as it fills, it is flushed automatically. It is almost certainly true that your buffer is flushed more frequently than the once-every-thousand PUT operations that you use. I would not explicitly FLUSH an INSERT cursor unless I needed to synchronize stuff on the disk. One issue to check would be whether COMMIT flushes the cursor. It should, since open cursors are closed at commit, and closing an INSERT cursor flushes it. >I believe there has to be a relationship between the size of the row in the >buffer, the number of rows in the buffer and writing to disk. Any ideas? There is indeed a connection between the size of row and the size of buffer, but it doesn't need to concern you very much. There is a connection to writing to the engine, which ultimately writes to disk, but the connection with writing to disk is minimal, precisley because the Engine, as separate program, does it. >What role does the checkpoint play? None. >Our tbconfig values for buffers are: These are not relevant either. Yours, Jonathan Leffler (johnl@informix.com) #include <disclaimer.h>