Hi, >In this case also, since we dont have eviction in place, all the time data >is retrieved from RAM only, the only time request goes to Oracle is for >Upserts and delete. Here, all queries run on data in RAM only. Ignite just propagate updates to back store (to Oracle via CacheStore impl) with keeping consistency guaranties according to CacheStore configuration.
>So if oracle DB is loaded heavily while running these queries also, it does >not affect the cluster performance when it comes to data retrievals, am i >right ?? Not quite, CacheStore affects performance due to either CacheStore updates are synchronous or CacheStore buffer is limited. With Ignite persistence you are not limited with RAM. PageMemory concept allow you to query data that resides on disk. What OOM do you mean Ignite or JVM, and on what side: client or server? Why you think with same dataset OOM with persistence has higher probability than with no persistence and no eviction? On Mon, Apr 2, 2018 at 9:09 AM, Naveen <naveen.band...@gmail.com> wrote: > HI > > Let me rephrase my question, guess I have conveyed my question correctly. > > Lets take an example > > Ignite cluster with backing store as RDBMS - oracle and no eviction in > place. > > As we all know, we can run complex queries on Oracle to retrieve desired > data. > In this case also, since we dont have eviction in place, all the time data > is retrieved from RAM only, the only time request goes to Oracle is for > Upserts and delete. > So if oracle DB is loaded heavily while running these queries also, it does > not affect the cluster performance when it comes to data retrievals, am i > right ?? > Inserts/Update/Deletes on DB may get slower if DB is loaded with these > heavy > queries. > > Similar way, how can we achieve this with ignite cluster with native > persistence ? > When native persistence is used, if we run some heavy queries on cluster > thru SQLLINE, it may give out of memory error and node might crash also. > How > can we avoid this, it should run on backing store, I am fine if thge query > exection takes longer time, but cluster should not get crashed. > > Hope I conveyed my requirements more clear this time. > > Thanks > Naveen > > > > -- > Sent from: http://apache-ignite-users.70518.x6.nabble.com/ > -- Best regards, Andrey V. Mashenkov