I didn't tell you the machine information. It's Itanium 2 with 2 processors
on each machine.
vendor : GenuineIntel
arch : IA-64
family : Itanium 2
model : 1
revision : 5
archrev : 0
features : branchlong
cpu number : 0
cpu regs : 4
cpu MHz : 1500.000000
itc MHz : 1500.000000
After the change of all bc_*** parameters. The code works fine for 2 nodes.
But the time scale between
2 and 4 nodes are almost the same. The list of the time difference bellow
shows that the problem is in
the force adjustment calculation. Is there a way to overcome this problem?
For 4 nodes
| Build the list
0.02 ( 2.09% of List )
|
Other
1.00 (97.91% of List )
|
List time
1.02 ( 0.41% of Nonbo)
|
Short_ene time
99.91 (95.31% of Direc)
|
Other
4.92 ( 4.69% of
Direc)
| Direct Ewald time 104.83 (42.80% of Ewald)
|
Force Adjust
113.30 (46.25% of Ewald)
|
Virial junk
22.12 ( 9.03% of Ewald)
| Start sycnronization 1.52 ( 0.62% of Ewald)
|
Other
3.18 ( 1.30% of Ewald)
| Ewald time 244.96 (99.01% of Nonbo)
|
Other
1.42 ( 0.58% of
Nonbo)
| Nonbond force 247.40 (74.52% of Force)
| Bond/Angle/Dihedral 25.82 ( 7.78% of Force)
| FRC Collect time 49.11 (14.79% of Force)
|
Other
9.66 ( 2.91% of Force)
| Force time 331.99 (84.39% of Runmd)
| Verlet update time 24.57 ( 6.25% of Runmd)
| CRD distribute time 32.69 ( 8.31% of Runmd)
| Other 4.14 ( 1.05% of Runmd)
| Runmd Time 393.39 (100.0% of Total)
| Other 0.09 ( 0.02% of Total)
| Total time 393.48 (100.0% of ALL )
For 2 nodes
|
Build the list
0.02 ( 2.15% of List )
|
Other
1.07 (97.85% of List )
|
List time
1.09 ( 0.43% of Nonbo)
|
Short_ene time
198.41 (98.22% of Direc)
|
Other
3.60 ( 1.78% of Direc)
| Direct Ewald time 202.00 (79.84% of Ewald)
|
Force Adjust
36.96 (14.61% of Ewald)
|
Virial junk
8.76 ( 3.46% of
Ewald)
| Start sycnronization 1.50 ( 0.59% of Ewald)
|
Other
3.80 ( 1.50% of Ewald)
| Ewald time 253.03 (98.93% of Nonbo)
|
Other
1.65 ( 0.64% of
Nonbo)
| Nonbond force 255.77 (79.07% of Force)
| Bond/Angle/Dihedral 47.91 (14.81% of Force)
| FRC Collect time 13.89 ( 4.30% of Force)
|
Other
5.91 ( 1.83% of Force)
| Force time 323.49 (92.61% of Runmd)
| Verlet update time 12.60 ( 3.61% of Runmd)
| CRD distribute time 8.64 ( 2.47% of Runmd)
| Other 4.57 ( 1.31% of Runmd)
| Runmd Time 349.29 (100.0% of Total)
| Other 0.04 ( 0.01% of Total)
| Total time 349.34 (100.0% of ALL )
Best regards
Qiang
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Thu Dec 02 2004 - 21:53:00 PST
