Qingxin's chart shows the performance difference clearly. In most case, insertTablet > insertRecords > insertRecord > JDBC.
----------------------------------- Xiangdong Huang School of Software, Tsinghua University 黄向东 清华大学 软件学院 冯 庆新 <qingxin.f...@hotmail.com> 于2022年6月6日周一 08:57写道： > Hello Trevor Hart > > Can you access this web link: > http://111.202.73.147:13000/d/5ZvuEYE7z/atm-biao-zhun-da-qi-ya-huan-jing?orgId=1&viewPanel=26 > Is this one ok? > > Config items in benchmark: > DB_SWITCH=IoTDB-013-JDBC / SESSION_BY_TABLET / SESSION_BY_RECORDS / > SESSION_BY_RECORD > GROUP_NUMBER=10 > LOOP=1000 > DEVICE_NUMBER=50 > SENSOR_NUMBER=500 > BATCH_SIZE_PER_WRITE=100 > POINT_STEP=1000 > OP_INTERVAL=0 > IS_OUT_OF_ORDER=false > > B.R > qingxin.feng > > > 发件人: Trevor Hart<mailto:tre...@ope.nz> > 发送时间: 2022年6月6日 8:46 > 收件人: dev<mailto:dev@iotdb.apache.org> > 主题: JDBC vs Java API > > Hello Team > > > > Does anyone have any published benchmark results of JDBC vs the Java API? > > > > Firstly Im aware of https://github.com/thulab/iotdb-benchmark but I dont > see any published results for the various API methods. > > > > I currently use JDBC for my non-realtime ingestion of data and while Ive > never encountered any bottle necks I am aware that the documentation says > that JDBC is not recommended for high velocity data. > > > > Ive done some very basic ingestions benchmarking tests of inserting 1 > million rows and the Java API is around 2x faster. Is this the typical > improvement between JDBC and Java API? > > > > For my simplistic test I am inserting 1 millions rows of timestamp & > incrementing row id eg insert into root.sg1.d1(timestamp,s1) > values(${DateTime.Now}, ${n}) > > > > With JDBC I get around 6000 rows per second. > > > > With the Java native API I get around 12000 rows per second using > session.executeNonQueryStatement. > > > > I assume insertTablets and insertRecord(s) would be even faster? > > > > Thanks > > Trevor Hart > >
