计算化学公社

标题: How to build a cluster ? [打印本页]

作者
Author: aitouna 时间: 2019-12-20 03:16
标题: How to build a cluster ?
本帖最后由 aitouna 于 2019-12-20 04:47 编辑

Dear everybody in the forum,
I have 5 Desktop with the configuration:
*Processor : Intel(R) Core(TM) i5-6600 CPU @ 3.30GHz
* RAM: 4,00 Go
All using windows 8.1 system.
I run my calculations at the different Desktop, but it take a very long time to complete.
I would like to learn how to build a cluster system, in purpose to save time. I watched a lot of tutorials but i didn't succeed to build it.
Please i need practical and easy method.

Sincerely,

作者
Author: fineren 时间: 2019-12-20 09:18
这也是国际化的论坛了

作者
Author: liyuanhe211 时间: 2019-12-20 09:32
本帖最后由 liyuanhe211 于 2019-12-20 16:26 编辑

If you mean cross-node parallelization, you might need extra hardware for cross-node communication to achieve satisfying efficiency. And if you use Gaussian, you need to buy the version of Gaussian that supports Linda, which is pretty expensive. Considering that your node is quite small, it feels pretty wasteful.

If you just want to build a cluster for simpler managing (like adding a queue system to it) and run single-node missions, it is quite simple. Just install CentOS on each node, configure the Linux version softwares, connect them to a router and install a queue managing software like SGE, PBS (there are plenty of tutorials online).

作者
Author: ghifi37 时间: 2019-12-20 10:43
本帖最后由 ghifi37 于 2019-12-20 10:44 编辑

To speed up the calculation, a cross-node computing resolution needed.
Any suggestion is based on your goal, like what is the software you will use and so on.

作者
Author: aitouna 时间: 2019-12-20 20:02
本帖最后由 aitouna 于 2019-12-21 01:56 编辑

I use gaussian 09 for my calculation, and i would like to speed up the calculation. I really want to learn how to built it for my personal curiosity of knowledge, and i have to optimize a big molecule containing transition metal, which will take a very long time.

作者
Author: pyscf 时间: 2019-12-21 06:42

fineren 发表于 2019-12-20 09:18
这也是国际化的论坛了

别人只是练练英语

作者
Author: sobereva 时间: 2019-12-22 16:09

pyscf 发表于 2019-12-21 06:42
别人只是练练英语

当前帖子是真实的外国人发的
也有不少外国人通过google翻译在本论坛看帖，比如http://bbs.keinsci.com/thread-9624-1-1.html

作者
Author: sobereva 时间: 2019-12-22 16:13

aitouna 发表于 2019-12-20 20:02
I use gaussian 09 for my calculation, and i would like to speed up the calculation. I really want to ...

Compared to struggling to build a cluster and run Gaussian, a much better solution is using ORCA code instead. The B97-3c method in ORCA is quite fast and reliable for systems containing transition metals.

作者
Author: abin 时间: 2020-1-2 14:27
How to build a cluster for computational chemistry, check out this link,
https://www-users.cs.york.ac.uk/ ... eowulf_cluster.html

Any way you should be familiar with Linux command line. Otherwise, any tutorial is a piece of shit.

By the way, several years before, I tried to run G09/Linda on PC_cluster which is built by two P4 3.0 desktop, linked by 100 bps switch.
G09 runs much faster.

作者
Author: aitouna 时间: 2020-1-2 20:59
Dear abin,
Thank you very much for sharing thos helpful information.

Sincerely,

欢迎光临计算化学公社 (http://bbs.keinsci.com/)