Thx, very helpfull reply. One more question: is it need to do "END"
after "BEGIN" or enought "COMMIT"?
By the way, some edition of sql query improve productivity a little bit too:
// Columns in 'data' table defined like (INTEGER, INTEGER, INTEGER, TEXT)
char * sql = sqlite3_mprintf( "INSERT INTO data VALUES( '%q', '%q',
'%q', '%q' )",
m_sources_map[ it->m_source ].m_sid.c_str(),
sec.c_str(), usec.c_str(), it->m_value.c_str() );
was changed with:
// Columns in 'data' table defined like (INTEGER, INTEGER, INTEGER, TEXT)
char * sql = sqlite3_mprintf( "INSERT INTO data VALUES( %d, %d, %d, '%s'
)",
m_sources_map[ it->m_source ].m_sid,
it->m_time.sec(), it->m_time.usec(), it->m_value.c_str() );
A common issue of high latency transactions. SQLite has a high
per-transaction overhead, which can be amortized across multiple
INSERTs or UPDATEs to improve the average INSERT rate. You are doing a
single INSERT per transaction, so wrap multiple INSERTs inside a
single "BEGIN" ... "END" transaction.
See:
http://www.sqlite.org/cvstrac/wiki?p=PerformanceConsiderations
Christian
--
Regards,
Igor Mironchick,
Intervale ©
#ICQ 492-597-570
-----------------------------------------------------------------------------
To unsubscribe, send email to [EMAIL PROTECTED]
-----------------------------------------------------------------------------
