Yakov, I've checked the current master and found out what only 2 of 4 system caches were updated to have copyOnRead=false. I'm wondering why.
The work was done for https://issues.apache.org/jira/browse/IGNITE-3569 issue. Regarding the hint I've filed the new issue: https://issues.apache.org/jira/browse/IGNITE-3714 2016-08-18 18:34 GMT+03:00 Yakov Zhdanov <yzhda...@apache.org>: > 1. Agree. You can use org.apache.ignite.internal. > GridPerformanceSuggestions > to print this warning. > 2. Disagree. For me this brings a lot of confusion. > > Alex, can you please check that all internal caches used in Ignite > internals have this property set to false. > > --Yakov > > 2016-08-18 17:24 GMT+03:00 Alexei Scherbakov <alexey.scherbak...@gmail.com > >: > > > Guys, > > > > Currently Ignite tries to be compatible with JSR-107 specification which > is > > "store-by-value" by default. > > > > So we have the copyOnRead flag with default value set to true in > > CacheConfiguration. > > > > The truth is the performance of Ignite with this flag disabled is > > significantly better. > > With a simple test reading locally 1Kb values I see about 7x performance > > gain with copying disabled. > > > > Then the https://issues.apache.org/jira/browse/IGNITE-2795 will be > > finished > > total performance impact will even be worse. > > > > I propose to add new performance hint related to copyOnRead property. > > > > It can be done in two ways: > > > > 1. Read static cache configurations on start and print hint if at least > one > > cache has copyOnRead=true. > > If the dynamic cache with copyOnRead=true is started print hint as well. > > > > 2. Introduce IgniteConfiguration.copyOnRead property, which is true by > > default. > > All caches will have their respective property set by default using > > IgniteConfiguration default value. > > If the IgniteConfiguration.copyOnRead=true when print hint on start. > > > > What do you think ? > > > > One more thing to mention. > > MutableConfiguration has storeByValue property, which has the same > meaning > > as copyOnRead, but strangely not used. > > This might be confusing for users expecting full JSR-107 compatibility. > > That's the reason for it? > > > > -- > > > > Best regards, > > Alexei Scherbakov > > > -- Best regards, Alexei Scherbakov
