That is my observation that mmap allocates N*4GB. However I don't know
how to verify. Assume I have 8 benchmarks. Simulating a uniprocessor
with 4GB memory with each of the benchmarks has no problem. I mean
-n 1 -b app1
-n 1 -b app2
-n 1 -b app3
...
-n 1 -b app8

However when I define 4GB and run
-n 8 -b app1,app2, ..., app8

the unmapped address is outside 4GB.

On 2/2/13, Ali Saidi <[email protected]> wrote:
> Why should it be N*4GB? In each process using 4GB of memory? That doesn't
> seem right. I'd guess your issue is that a bad request is being generated
> somewhere in the memory system and making it incredibly large is just
> covering the problem up.
>
> Ali
>
> On Feb 2, 2013, at 2:57 AM, Mahmood Naderan <[email protected]> wrote:
>
>> It is really annoying. I think mmap is responsible for this behavior.
>>
>> On 1/31/13, Mahmood Naderan <[email protected]> wrote:
>>> Hi
>>> I have found that in a N multicore simulation, defined memory size
>>> should be N*4GB. For example, I am simulting eight cores so I have to
>>> define 32GB memory. Otherwise I randomly get "unable to find
>>> destination ..." error.
>>>
>>> However in reality while the simulation is running, "top" command
>>> shows that about 1.5GB of memory is used.
>>>
>>> Another problem is that in a systeam with 32GB memory installed and
>>> 20GB free memory, I can not define a 32GB memory in gem5 simulation
>>> script. Otherwise I get "could not mmap" error message.
>>>
>>> Is there any better memory management?
>>>
>>>
>>> Regards,
>>> Mahmood
>>>
>>
>>
>> --
>> Regards,
>> Mahmood
>> _______________________________________________
>> gem5-users mailing list
>> [email protected]
>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>>
>
> _______________________________________________
> gem5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>


-- 
Regards,
Mahmood
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to