[ https://issues.apache.org/jira/browse/IGNITE-7531?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16361305#comment-16361305 ]
Pavel Kuznetsov commented on IGNITE-7531: ----------------------------------------- Implemented 2 benchmarks: sequence of single insert and batched insert current results : insert 10^6 tuples (long id, 5 long fields and 5 string fields), 1 driver machine, 1 server, 1 client (thin driver connects to) warmed up with 3*10^6 inserts single insert : 225 sec batching 10 : 53 sec batching 50 : 31 sec batching 100 : 31 sec batching 1000 : 30 sec batching 10^4 : 27 sec > SQL: Create data load benchmarks > -------------------------------- > > Key: IGNITE-7531 > URL: https://issues.apache.org/jira/browse/IGNITE-7531 > Project: Ignite > Issue Type: Task > Components: sql, yardstick > Reporter: Vladimir Ozerov > Assignee: Pavel Kuznetsov > Priority: Major > Fix For: 2.5 > > > We need to implement a set of data loading benchmarks to better understand > how fast Ignite is able to consume data. This task consists of two steps: > 1) Extend Yardstick capabilities > 2) Create set of benchmarks > 1) Yardstick > Data load benchmark should be executed in single-shot mode: only one > iteration, only total execution time is needed, start callback for setup and > warmup, stop callback for cleanup. > Currently Yardstick cannot do that, so we need to extend it. Possibly, we can > control this through new {{boolean BenchmarkDriver.isSingleShot()}} method. > 2) Benchmarks > At first let's focus on thin JDBC driver. The following cases should be > executed: > 2.1) Normal INSERT > 2.2) Batched INSERT > 2.3) Streaming INSERT (when IGNITE-7253 is ready) > 2.4) P. 1-3 with and without dynamically disabled WAL (ALTER TABLE ... > NOLOGGING) > 2.5) P. 1-3 with additional indexes - either created before data load on > empty table, or after load on table with data. -- This message was sent by Atlassian JIRA (v7.6.3#76005)