It was on JDK 1.6.0_u16 client compiler. Persisting about 2 000 "large" entities (~45 attributes) and 200 000 entities (~4 attributes) in one transaction (tuned with: batch writes, cache all statements, batch size = 200, bind all parameters) Profiler: JProfiler version 4.2.4 I'm not "JDK" guru so I'm not able to make a patch. I just found that there is used RegisterNatives in Object.hashCode, but not in System.identityHashCode.
Thank you. ----- Original Message ----- From: "Christian Thalinger" <[email protected]> To: "Tom Hawtin" <[email protected]> Cc: "Janda Martin" <[email protected]>, [email protected] Sent: Tuesday, September 22, 2009 1:47:44 PM GMT +01:00 Amsterdam / Berlin / Bern / Rome / Stockholm / Vienna Subject: Re: System.identityHashCode performance Tom Hawtin wrote: > Janda Martin wrote: >> I would like to ask about any plans of speed improvements to >> "System.identityHashCode" or "IdentityHashMap" >> >> I use EclipseLink and it uses IdentityHashMap for storing entities and their >> clones. >> >> In my case System.identityHashCode consumes 54% of time during insert data >> to database. There are lots of other project using IdentityHashMap and >> identityHashCode. > > CR 6378256 seems to cover this for the client/C1 compiler. Presumably > the server/C2 compiler already intrinsifies the method. It's been in > "fix understood" for a couple of years... > > http://bugs.sun.com/view_bug.do?bug_id=6378256 Martin, are you having performance problems with client or server compiler? -- Christian
