Ok I think I have managed to reproduce the issue. Here is the github code: https://github.com/javadevmtl/ignite-for-github
The code is simple it launches a FixedThreadPool each worker executes a put and SQL query. The app can be launched in on-heap or off-heap mode The average time is calculated by: total time of execution / number of executions. on-heap 1 Million Entries: Avg 128128 nanos 2 Million Entries: Avg 141960 nanos 3 Million Entries: Avg 149321 nanos 4 Million Entries: Avg 167495 nanos off-heap 1 Million Entries: Avg 333946 nanos 2 Million Entries: Avg 433551 nanos 3 Million Entries: Avg 453551 nanos 4 Million Entries: Avg 546028 nanos As you can see on-heap is allot faster and far more stable in latency time. Off-heap seems to get slower and slower with the cache growing. The app is run as: java -jar -Xmx16g -XX:+UseG1GC -jar ignite-for-github-1.0-fat.jar on-heap 8 4000000 1000 130 Only the off-heap/on-heap flag and the number of entries are changed between runs... -- View this message in context: http://apache-ignite-users.70518.x6.nabble.com/SQL-Performance-indexing-performance-on-heap-vs-off-heap-tp1352p1364.html Sent from the Apache Ignite Users mailing list archive at Nabble.com.
