I don't know this problem, maybe you could send more information about what
you are doing. I have used 32 replicas without trouble. can you do normal MD
with that many nodes to test them? is there any error message in the sander
outputs?
On 2/20/07, Seongeun Yang <seongeun.korea.ac.kr> wrote:
>
> Hello all,
>
> I met a problem in running REMD.
>
> I'm using amber8 installed on Intel Xeon clusters (dual core).
>
> After I found that test REMD runs using 16 replicas on 4 nodes were
> successful for at least 1000 exchanges,
> I tried to run the same job using 32 replicas on 8 nodes.
>
> After just 4 exchanges, the job stopped giving the following error
> messages.
>
> These are a few lines at the beginning of the error messages.
>
> .....
> *** glibc detected *** double free or corruption (!prev):
> 0x00000000016f2600 ***
> p23_4663: p4_error: interrupt SIGx: 6
> rm_l_23_4676: (405.574219) net_send: could not write to fd=5, errno = 32
> p1_6052: p4_error: net_recv read: probable EOF on socket: 1
> p2_6069: p4_error: net_recv read: probable EOF on socket: 1
> rm_l_1_6065: (407.609375) net_send: could not write to fd=5, errno = 32
> p3_6086: p4_error: net_recv read: probable EOF on socket: 1
> rm_l_2_6082: (407.582031) net_send: could not write to fd=5, errno = 32
> p28_3283: p4_error: net_recv read: probable EOF on socket: 1
> rm_l_28_3297: (404.835938) net_send: could not write to fd=5, errno = 32
> .....
>
> Please let me know what is the source of the error in this case.
>
> Thanks for your answers in advance.
>
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Wed Feb 21 2007 - 06:07:41 PST
