Hello! So I was increasing amount of RAM in the memory model, and it turns out that Off-Heap usage will not grow past:
2018-09-11 12:47:46,603 INFO [pub-#290] log4j.Log4JLogger (Log4JLogger.java:566) - ############################################# 2018-09-11 12:47:56,605 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - Show metrics inside ignite c2785f18-983c-490e-8ebc-3198b54ae132 2018-09-11 12:47:56,606 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - Size : 100000 of cache contactsEx 2018-09-11 12:47:56,606 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - ############################################# 2018-09-11 12:47:56,606 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - >>> Memory Region Name: Default_Region 2018-09-11 12:47:56,606 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - AllocationRate: 0.0 2018-09-11 12:47:56,606 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - PagesFillFactor: 0.8341349 2018-09-11 12:47:56,606 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - PhysicalMemoryPages: 35493 2018-09-11 12:47:56,606 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - OffHeapSize: 209715200 2018-09-11 12:47:56,606 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - CheckpointBufferSize: 0 2018-09-11 12:47:56,606 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - CheckpointBufferPages: 0 2018-09-11 12:47:56,607 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - OffheapUsedSize: 145379328 2018-09-11 12:47:56,607 INFO [pub-#292] log4j.Log4JLogger (Log4JLogger.java:566) - ############################################# After half an hour usage is still at this point. So I imagine that storing this amount of data in Apache Ignite takes 140M and not less. But it won't grow past this point. There may be a lot of reasons why this number may grow before it reaches plateau. There's, obviously, a lot of metadata pages, some of which may not be allocated immediately. Then there's fragmentation: If you remove an object from page, and write a slightly larger object, it may not fit the page and it will use up space on some other page. PagesFillFactor is metric that chases this. Note that for such a small cache, PDS will be absolutely dominated by metadata. On large datasets you will see growth due to fragmentation. But neither of those are runaway growth. Unfortunately, your reproducer does not show runaway growth either, so I can't tell you anything further. Regards, -- Ilya Kasnacheev вт, 11 сент. 2018 г. в 11:28, Serg <sergey.ma...@ardas.dp.ua>: > Hi Ilya, > > I created reproducer with two tests > https://github.com/SergeyMagid/ignite-reproduce-grow-memory > > Differents in this tests only is data which inserted to cache. > I have previously suppose that problem caused with BinaryObject only but I > reproduced this problem without BinaryObject too. > > > > -- > Sent from: http://apache-ignite-users.70518.x6.nabble.com/ >