|

»

»

»

»

Memory allocation problem?

View: New views

3 Messages — Rating Filter: Alert me

	Memory allocation problem? by Eunseok, Kim :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message Some parts of this message have been removed. Learn more about Nabble's security policy. Hi~ I’m James Kim again. I increased VmWare memory to 1024M, then the allocation of matrix was successful. ‘kaffe –Xmx 1024m Benchmark 10000 1’ works well. Why does sablevm require larger memory than other jvm? Is this related memory option? I still don’t know why sablevm does not work in 256M memory. If anybody know it, then please let me know~ Have a nice day~ Thanks~ James _______________________________________________ SableVM-user mailing list SableVM-user@... http://sablevm.org/lists/control/listinfo/sablevm-user
	Re: Memory allocation problem? by Etienne Gagnon :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message Hi James, SableVM uses a stop/copy collector. This has advantages and drawbacks. Main advantages: - It is a precise collector. All "unreachable" objects are collected. This is not always the case in Kaffe, for example. - It doesn't suffer any Java heap memory fragmentation. This means that a very long running process won't endup being unable to use available memory because it is too fragmented. Main disadvantage: - It only uses 50% of heap space, as it uses the other 50% as a target while copying objects during collection. - It is not a hybrid/generational collector as Sun's one. Now, SableVM can be compiled to use Carl Lebsack's generational collector. Unless I'm wrong, I think that all generations are collected using a copying collector, so this won't necessarily solve your tight space problem. I'll let Carl discuss the matter, if you have additional questions on this collector. I am puzzled, though... Don't you have more "virtual memory" than physical memory (i.e. RAM)? Usually, applications are limited by the amount of available virtual space, not by RAM. Etienne Eunseok, Kim wrote: > I increased VmWare memory to 1024M, then the allocation of matrix was > successful. > > ‘ kaffe – Xmx 1024m Benchmark 10000 1 ’ works well. > > Why does sablevm require larger memory than other jvm? > > Is this related memory option? > > I still don ’ t know why sablevm does not work in 256M memory. > > If anybody know it, then please let me know~ -- Etienne M. Gagnon, Ph.D. http://www.info2.uqam.ca/~egagnon/ SableVM: http://www.sablevm.org/ SableCC: http://www.sablecc.org/ _______________________________________________ SableVM-user mailing list SableVM-user@... http://sablevm.org/lists/control/listinfo/sablevm-user signature.asc (264 bytes) Download Attachment
	Re: Memory allocation problem? by Carl Lebsack :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message On 6/1/06, Etienne Gagnon <egagnon@...> wrote: > Hi James, > > SableVM uses a stop/copy collector. This has advantages and drawbacks. > > Main advantages: > - It is a precise collector. All "unreachable" objects are collected. > This is not always the case in Kaffe, for example. > - It doesn't suffer any Java heap memory fragmentation. This means that > a very long running process won't endup being unable to use available > memory because it is too fragmented. > > Main disadvantage: > - It only uses 50% of heap space, as it uses the other 50% as a target > while copying objects during collection. > - It is not a hybrid/generational collector as Sun's one. > > Now, SableVM can be compiled to use Carl Lebsack's generational > collector. Unless I'm wrong, I think that all generations are collected > using a copying collector, so this won't necessarily solve your tight > space problem. I'll let Carl discuss the matter, if you have additional > questions on this collector. > Etienne is correct. The generational collector still employs copying for the mature space (majority of the heap) and will not provide substantially more allocation space than the original copying collector. No VM should be able to run your example with a 10,000x10,000 matrix in 256MB as the matrix alone would be ~400MB. Kaffe uses a mark-sweep collector and should be able to run with slightly larger than the 400MB (slightly is not a precise term ;-) ). SableVM should require more than 800MB to run this test case due to the copying collector (only half the heap is used at a time). The generational variant will provide no additional advantage for this benchmark. What exactly are you trying to measure with your benchmark? Carl Lebsack > I am puzzled, though... Don't you have more "virtual memory" than > physical memory (i.e. RAM)? Usually, applications are limited by the > amount of available virtual space, not by RAM. > > Etienne > > Eunseok, Kim wrote: > > I increased VmWare memory to 1024M, then the allocation of matrix was > > successful. > > > > ' kaffe – Xmx 1024m Benchmark 10000 1 ' works well. > > > > Why does sablevm require larger memory than other jvm? > > > > Is this related memory option? > > > > I still don ' t know why sablevm does not work in 256M memory. > > > > If anybody know it, then please let me know~ > > -- > Etienne M. Gagnon, Ph.D. http://www.info2.uqam.ca/~egagnon/ > SableVM: http://www.sablevm.org/ > SableCC: http://www.sablecc.org/ > > > > _______________________________________________ > SableVM-user mailing list > SableVM-user@... > http://sablevm.org/lists/control/listinfo/sablevm-user > > > > _______________________________________________ SableVM-user mailing list SableVM-user@... http://sablevm.org/lists/control/listinfo/sablevm-user

LightInTheBox - Buy quality products at wholesale price!