d*****w 发帖数: 124 | 1 我们目前拥有两套用于分子模拟的Beowulf计算机群。一套是用于教学目的8 CPU
AMD Athlon MP 2000+ (主频1666M Hz) 构架的Beowulf计算机群,用100M Ethernet连接
,代号为ding,搭建于2002年10月。另一套是上个月搭建的用于多相多组份高分子体系模
拟的机群,代号为”m”,拥有46个Intel Xeon 2.4 G CPU,以1000M Ethernet连接。
为了便于和其他Beowulf计算机群进行性能比较,最近我们用通用的Linpack软件包对这两
个系统进行的测试。下面是整个测试过程和结果的报告:
http://pplas.icas.ac.cn/~qiliao/cluster.htm
http://pplas.icas.ac.cn/~qiliao/beowulf/Linpack_benchmark.htm |
b**s 发帖数: 589 | 2 I got a question,
how many chain segments do you think is enough for the model to simulate the
real polymer chain?
接
模
两
【在 d*****w 的大作中提到】![](/moin_static193/solenoid/img/up.png) : 我们目前拥有两套用于分子模拟的Beowulf计算机群。一套是用于教学目的8 CPU : AMD Athlon MP 2000+ (主频1666M Hz) 构架的Beowulf计算机群,用100M Ethernet连接 : ,代号为ding,搭建于2002年10月。另一套是上个月搭建的用于多相多组份高分子体系模 : 拟的机群,代号为”m”,拥有46个Intel Xeon 2.4 G CPU,以1000M Ethernet连接。 : 为了便于和其他Beowulf计算机群进行性能比较,最近我们用通用的Linpack软件包对这两 : 个系统进行的测试。下面是整个测试过程和结果的报告: : http://pplas.icas.ac.cn/~qiliao/cluster.htm : http://pplas.icas.ac.cn/~qiliao/beowulf/Linpack_benchmark.htm
|
w********h 发帖数: 12367 | 3 experimentally, only when N is larger than Gaussian chain length,
we can think it is a real polymer chain.
Ethernet连
系
。
这
【在 b**s 的大作中提到】![](/moin_static193/solenoid/img/up.png) : I got a question, : how many chain segments do you think is enough for the model to simulate the : real polymer chain? : : 接 : 模 : 两
|
v*****n 发帖数: 621 | |
b**s 发帖数: 589 | 5 how much is Gaussian chain length normally? could you give a example
the
体
接
对
【在 w********h 的大作中提到】![](/moin_static193/solenoid/img/up.png) : experimentally, only when N is larger than Gaussian chain length, : we can think it is a real polymer chain. : : Ethernet连 : 系 : 。 : 这
|
d*****w 发帖数: 124 | 6 I would like to refer the classical work of Kremer on the simulation of
polymer melts:
"Dynamics of entangled linear polymer melts: A molecular-dynamics simulation",
J. Chem. Phys.,
92,5057.
In the paper, he computationally "measured" the entanglement length of the
coarse-grained model using in the following decade by the diffussion constant.
The entanglement length is found to be around 35. Thus if you want to
simulation the real physics of polymer melts by this model, the chain length
should b
【在 b**s 的大作中提到】![](/moin_static193/solenoid/img/up.png) : I got a question, : how many chain segments do you think is enough for the model to simulate the : real polymer chain? : : 接 : 模 : 两
|
w********h 发帖数: 12367 | 7 I read that paper.
seems his group can do the maximum chain length simulation.
simulation",
constant.
and
the
体
接
对
【在 d*****w 的大作中提到】![](/moin_static193/solenoid/img/up.png) : I would like to refer the classical work of Kremer on the simulation of : polymer melts: : "Dynamics of entangled linear polymer melts: A molecular-dynamics simulation", : J. Chem. Phys., : 92,5057. : In the paper, he computationally "measured" the entanglement length of the : coarse-grained model using in the following decade by the diffussion constant. : The entanglement length is found to be around 35. Thus if you want to : simulation the real physics of polymer melts by this model, the chain length : should b
|
w********h 发帖数: 12367 | 8 different for different polymer species.
less than Ne (for entangled molecular weight).
that's a number to make Rouse model applicable.
子
CPU,以1000M Ethernet连
包
【在 b**s 的大作中提到】![](/moin_static193/solenoid/img/up.png) : how much is Gaussian chain length normally? could you give a example : : the : 体 : 接 : 对
|
c***r 发帖数: 4631 | 9
接
模
两
which compiler are you using? gcc or icc?
【在 d*****w 的大作中提到】![](/moin_static193/solenoid/img/up.png) : 我们目前拥有两套用于分子模拟的Beowulf计算机群。一套是用于教学目的8 CPU : AMD Athlon MP 2000+ (主频1666M Hz) 构架的Beowulf计算机群,用100M Ethernet连接 : ,代号为ding,搭建于2002年10月。另一套是上个月搭建的用于多相多组份高分子体系模 : 拟的机群,代号为”m”,拥有46个Intel Xeon 2.4 G CPU,以1000M Ethernet连接。 : 为了便于和其他Beowulf计算机群进行性能比较,最近我们用通用的Linpack软件包对这两 : 个系统进行的测试。下面是整个测试过程和结果的报告: : http://pplas.icas.ac.cn/~qiliao/cluster.htm : http://pplas.icas.ac.cn/~qiliao/beowulf/Linpack_benchmark.htm
|
d*****w 发帖数: 124 | 10 gcc 3.2.2
Intel C++ compiler costs 6000 yuan for one license in China and the list price
is 12,000 Yuan.
We could not afford that.
【在 c***r 的大作中提到】![](/moin_static193/solenoid/img/up.png) : : 接 : 模 : 两 : which compiler are you using? gcc or icc?
|
|
|
v*****n 发帖数: 621 | 11 用个盗版的?icc编译出来的好像要快一些的。
price
【在 d*****w 的大作中提到】![](/moin_static193/solenoid/img/up.png) : gcc 3.2.2 : Intel C++ compiler costs 6000 yuan for one license in China and the list price : is 12,000 Yuan. : We could not afford that. :
|
c***r 发帖数: 4631 | 12
price
I remember Intel will provide trial version for free, at least long time ago.
and you can try it see is there any difference.
To be honest, if you are feeling very disappointed with DING, don't expect icc
could improve very much, though gcc fortran compiler is no good. BTW, I don't
like AMD CPU, except those 64bit ones.
I think one of easy ways to improve performance fo DING is to tweak the core
of linux, especially the network portion and network hardware.
You can also refer top500's re
【在 d*****w 的大作中提到】![](/moin_static193/solenoid/img/up.png) : gcc 3.2.2 : Intel C++ compiler costs 6000 yuan for one license in China and the list price : is 12,000 Yuan. : We could not afford that. :
|
M*******n 发帖数: 508 | 13 I remember intel has free Fortran and C++ compiler for Linux user
【在 d*****w 的大作中提到】![](/moin_static193/solenoid/img/up.png) : gcc 3.2.2 : Intel C++ compiler costs 6000 yuan for one license in China and the list price : is 12,000 Yuan. : We could not afford that. :
|
c***r 发帖数: 4631 | 14
price
just checked it, still free.
【在 M*******n 的大作中提到】![](/moin_static193/solenoid/img/up.png) : I remember intel has free Fortran and C++ compiler for Linux user
|
d*****w 发帖数: 124 | 15 我弄混了。 C++ 是free,上面的价格是 fortran compiler 7.1 for linux
【在 c***r 的大作中提到】![](/moin_static193/solenoid/img/up.png) : : price : just checked it, still free.
|
Ag 发帖数: 481 | 16 I remember, fortran for linux is also free
but without support.
【在 d*****w 的大作中提到】![](/moin_static193/solenoid/img/up.png) : 我弄混了。 C++ 是free,上面的价格是 fortran compiler 7.1 for linux
|
c***r 发帖数: 4631 | 17 agreeing...
【在 Ag 的大作中提到】![](/moin_static193/solenoid/img/up.png) : I remember, fortran for linux is also free : but without support.
|
d*****w 发帖数: 124 | 18 it was befero ver 7.0. now it is not anymore.
see:
http://www.intel.com/software/products/compilers/flin/
【在 c***r 的大作中提到】![](/moin_static193/solenoid/img/up.png) : agreeing...
|
n*c 发帖数: 5 | |
d*****w 发帖数: 124 | |
c***r 发帖数: 4631 | 21 so the better choice is to optimize the g77. And improve network. I heard that
the Virginia Tech G5 X cluster reached 10T by optimizing network performance.
BTW, G5 beated Xeon on top500.
【在 d*****w 的大作中提到】![](/moin_static193/solenoid/img/up.png) : 看样子我差点被骗局:) : 我也是从那里下的, 整个过程有点搞笑: : http://pplas.icas.ac.cn/bbs/topic.cgi?forum=5&topic=80&show=0 : 我有点晕了.
|