As an aside (and up front before the other numbers), I cleaned up some code around the MetaData area so that ArrayLists were trimmed and the other methods getUniqueChild, etc called my new getChildArrayByTagName method. I also finalized methods in this area and Containers and took out variable declarations within loops in these areas. The results were: RT: 115 ms (no change), min. 30 ms, max 3405 ms, VSZ: 183516Kb (down 0.37%), RSS: 106756Kb (down 17.57%). I probably gave up any performance gains for cleaning memory on the spot. It was interesting the in-memory use dropped that much. But IBM take the opposite view - sacrifice memory for speed.
As to your question Bela, Trove actually provides benchmarks - the creators want to make sure they are creating tight implementations. So out of the box, you can run your own numbers. I'll work my way around to concurrency issues once I'm satisfied with how things work in the current set up. But looking at the JBoss code at the moment, if it doesn't have concurrency issues now, I don't think the Trove classes will have much of an impact here. On my Win2K, PIII 1.2GHz with Sun JDK: -------------------------------- java.vm.name=Java HotSpot(TM) Client VM java.runtime.version=1.4.1_01-b01 os.name=Windows 2000 os.arch=x86 os.version=5.0 -------------------------------- compares 100000 Set.contains() operations. 3 are actually present in set Iterations: 10 Their total (msec): 380 Our total (msec): 771 Their average (msec): 38 Our average (msec): 77 -------------------------------- sums a 100000 element Set of Integer objects. Their approach uses Iterator.hasN ext()/next(); ours uses THashSet.forEach(TObjectProcedure) Iterations: 10 Their total (msec): 380 Our total (msec): 130 Their average (msec): 38 Our average (msec): 13 -------------------------------- compares Iterator.hasNext()/ Iterator.next() over 100000 keys Iterations: 10 Their total (msec): 270 Our total (msec): 151 Their average (msec): 27 Our average (msec): 15 compares Iterator.next() over 100000 map keys Iterations: 10 Their total (msec): 241 Our total (msec): 90 Their average (msec): 24 Our average (msec): 9 -------------------------------- compares 100000 LinkedList.add() operations Iterations: 10 Their total (msec): 851 Our total (msec): 190 Their average (msec): 85 Our average (msec): 19 -------------------------------- 100000 entry primitive int map.put timing run; no basis for compariso Iterations: 10 Their total (msec): 0 Our total (msec): 311 Their average (msec): 0 Our average (msec): 31 -------------------------------- compares 100000 Map.put() operations Iterations: 10 Their total (msec): 2173 Our total (msec): 451 Their average (msec): 217 Our average (msec): 45 -------------------------------- compares 20000 Set.contains() operations Iterations: 10 Their total (msec): 130 Our total (msec): 70 Their average (msec): 13 Our average (msec): 7 -------------------------------- compares 100000 Map.get() operations Iterations: 10 Their total (msec): 470 Our total (msec): 371 Their average (msec): 47 Our average (msec): 37 -------------------------------- compares 100000 Set.add() operations Iterations: 10 Their total (msec): 1843 Our total (msec): 350 Their average (msec): 184 Our average (msec): 35 -------------------------------- done -------------------------- Compare size of Set implementation: 1,000 Integer objects measured in bytes javasoft: 48263 trove: 28856 trove's collection requires 59% of the memory needed by javasoft's collection -------------------------- Compare size of LinkedList implementation: 1,000 TLinkableAdaptor objects measur ed in bytes javasoft: 40048 trove: 16024 trove's collection requires 40% of the memory needed by javasoft's collection -------------------------- Compare size of int/IntegerArrayList implementation: 1,000 ints measured in byte s javasoft: 20040 trove: 4032 trove's collection requires 20% of the memory needed by javasoft's collection -------------------------- Compare size of Map implementation: 1,000 Integer->Integer mappings measured in bytes javasoft: 48248 trove: 41688 trove's collection requires 86% of the memory needed by javasoft's collection ************************************************************************** ***** On Linux RedHat 7.2, PII 500MHz with IBM SDK: -------------------------------- java.vm.name=Classic VM java.runtime.version=1.4.0 os.name=Linux os.arch=x86 os.version=2.4.18 -------------------------------- compares 100000 Set.contains() operations. 3 are actually present in set Iterations: 10 Their total (msec): 934 Our total (msec): 1989 Their average (msec): 93 Our average (msec): 198 -------------------------------- sums a 100000 element Set of Integer objects. Their approach uses Iterator.hasNext()/next(); ours uses THashSet.forEach(TObjectProcedure) Iterations: 10 Their total (msec): 705 Our total (msec): 227 Their average (msec): 70 Our average (msec): 22 -------------------------------- compares Iterator.hasNext()/ Iterator.next() over 100000 keys Iterations: 10 Their total (msec): 508 Our total (msec): 470 Their average (msec): 50 Our average (msec): 47 -------------------------------- compares Iterator.next() over 100000 map keys Iterations: 10 Their total (msec): 471 Our total (msec): 266 Their average (msec): 47 Our average (msec): 26 -------------------------------- compares 100000 LinkedList.add() operations Iterations: 10 Their total (msec): 796 Our total (msec): 626 Their average (msec): 79 Our average (msec): 62 -------------------------------- 100000 entry primitive int map.put timing run; no basis for comparison Iterations: 10 Their total (msec): 0 Our total (msec): 678 Their average (msec): 0 Our average (msec): 67 -------------------------------- compares 100000 Map.put() operations Iterations: 10 Their total (msec): 2321 Our total (msec): 911 Their average (msec): 232 Our average (msec): 91 -------------------------------- compares 20000 Set.contains() operations Iterations: 10 Their total (msec): 274 Our total (msec): 224 Their average (msec): 27 Our average (msec): 22 -------------------------------- compares 100000 Map.get() operations Iterations: 10 Their total (msec): 734 Our total (msec): 779 Their average (msec): 73 Our average (msec): 77 -------------------------------- compares 100000 Set.add() operations Iterations: 10 Their total (msec): 2233 Our total (msec): 706 Their average (msec): 223 Our average (msec): 70 -------------------------------- done -------------------------- Compare size of Set implementation: 1,000 Integer objects measured in bytes javasoft: 56217 trove: 25662 trove's collection requires 45% of the memory needed by javasoft's collection -------------------------- Compare size of LinkedList implementation: 1,000 TLinkableAdaptor objects measured in bytes javasoft: 44778 trove: 20759 trove's collection requires 46% of the memory needed by javasoft's collection -------------------------- Compare size of int/IntegerArrayList implementation: 1,000 ints measured in bytes javasoft: 18445 trove: 3438 trove's collection requires 18% of the memory needed by javasoft's collection -------------------------- Compare size of Map implementation: 1,000 Integer->Integer mappings measured in bytes javasoft: 54275 trove: 38409 trove's collection requires 70% of the memory needed by javasoft's collection As with any tests, you need to draw your own conclusions. JonB > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] Behalf Of Bela Ban > Sent: Tuesday, 1 July 2003 4:34 AM > To: [EMAIL PROTECTED] > Subject: Re: [JBoss-user] On the performance trail - again > > > Jon, > > do you have any numbers on this ? E.g. Sun's collection classes against > Trove ? > > Performance is a moving target; ie. in one release of the VM, object > creation kills you, and in the next it is gone (JDK 1.4). Or try-catch > is bad, then in 1.4. the overhead is gone too.
smime.p7s
Description: S/MIME cryptographic signature
