Description of the flags used for the Fujitsu SPECjbb2000 results: Fujitsu Ltd. Dec 6, 2005 < Hardware setting > Memory system is in "Non Mirror Mode". PRIMEQUEST 480 memory system supports DSSA (Dual Sync System Architecture ) and works in one of following two mode. (1) Mirror Mode Address buses and Data buses are duplicated. And most of internal action of chipset is also duplicated. In this mode, system memory throughput becomes half but higher reliability is expected by the memory system duplication. (2) Non Mirror Mode Address buses and data buses are not duplicated. And the internal action of chipset is not duplicated. In this mode, full memory bandwidth is available, but the system duplication for higher reliability does not work. Please see the detail of DSSA in the attached file. "Fujitsu-linux-20051026_2.pdf". < Policy setting > numactl --interleave=all Interleave memory allocations on a set of nodes --cpubind=nodemask Execute the process only on the CPUs of the nodes specified in nodemask. It can be specified in addition to other policies because it affects the scheduler separately. --localalloc Sets the process policy to the standard local allocation policy. < JVM Command Line Options > -Xms Sets the initial size of the heap. This value can be specified in kilobytes (K,k), megabytes (M,m), or gigabytes (G,g). Default: The default value depends upon the JVM mode: -server mode: The default value is the lesser of 25% of the free physical memory or 64 MB. -client mode: The default value is the lesser of 25% of the free physical memory or 16 MB. -Xmx Sets the maximum size of the heap. This value can be specified in kilobytes (K,k), megabytes (M,m), or gigabytes (G,g). Use the following guidelines to determine this value: Default: The default value is the lesser of 75% of the total physical memory or 1 GB. -XXaggressive Performs adaptive optimizations as early as possible in the Java application run. To do this, the bottleneck detector will run with a higher frequency from the start and then gradually lower its frequency. This options also tells BEA JRockit to use the available memory aggressively. -XXtlasize=2m This flag specifies how large a Thread Local Area (TLA) should be. A TLA is a chunk of memory on the heap where a thread creates objects. The thread has exclusive use on this area and therefore object allocations are very cheap inside the TLA. -XXlargeObjectLimit=128k This flag specifies the limit for when objects are treated as large objects. This value can be specified in kilobytes (K,k), megabytes (M,m), or gigabytes (G,g). Default: The default value is 2k. -XXallocClearChunks This option allows you to clear a TLA for references and values at TLA allocation time and pre-fetch the next chunk. When an integer, a reference, or anything else is declared, it has a default value of 0 or null (depending upon type). At the appropriate time, you will need to clear these references and values to free the memory on the heap so Java can use-or reuse-it. You can do either when the object is allocated or, by using this option, when you request a new TLA. Default: By default, this mechanism is disabled. If you use this option but do not specify a boolean value, the default is true.