OpenFOAM on Amazon EC2
 

Cluster compute instances announced...
http://aws.amazon.com/ec2/instance-types/

I hope to be able to share an openfoam-extend machine image and instructions of how to sign up and start using it on the wiki before too long.

Questions, comments, recommendations much appreciated.

Regards,
Ben Racine

And the AMI is?

You can try the CAELinux image in EC2, it contains openfoam 1.7 and many other software packages like Discretizer, enGrid, netgen, Salome, GMSH
see http://www.caelinux.com for more details...

If there is a gpu card we could accelerate OF simulations. We have already good results with simplefoam. See http://sourceforge.net/projects/openfoamspeedit/ for details or drop me a line.

lukazs, you are annoying me and coworkers more and more .you spam in many threads to advertise a commercial offer even if nobody has asked anything remotely related .you even ask people to leave the discussion on this forum and use your site instead .

i remind you of the forum rules that you have agreed :

"Link spammers and other forms of spam are hunted very actively, both automatically and manually."

• "State your affiliation explicitly in your message if you:
  • Recommend a product sold by your employer"
• "Be very restrictive in posting links to other web sites, especially commercial sites"

"Forbidden content:

• Advertising, both commercial and personal"

i ask you to stop your practice .
without regards ,zhijun !

I agree with zhijun. We should not provide links to our web pages. Apologies for that. Our intention was to share our GPL-based technology to OF users, and not to advertise it on the forum.

Therefore, from now on we will stop providing a link to our web page but only to sourceforge where the OF plugin and the opensource speedit library can be downloaded: http://sourceforge.net/projects/openfoamspeedit/

Apologies again to those who felt offended.

Best regards,
Lukasz

Here they are :

'us-east-1' :

• 'openfoam171_0-1ubuntu2' - 'ami-ecf50385',
• 'openfoam171_0-1' - 'ami-62fa0c0b',
• 'openfoam-dev-1.5' - 'ami-98f701f1',
• 'openfoam-1.6-ext' : 'ami-b2fd0cdb',

'eu-west-1' :

• 'openfoam171_0-1ubuntu2' - 'ami-519ca925',
• 'openfoam171_0-1' - 'ami-8d9da8f9',
• 'openfoam-dev-1.5' - 'ami-4d9ca939',
• 'openfoam-1.6-ext' - 'ami-f998ad8d',

'ap-southeast-1' :

• 'openfoam171_0-1ubuntu2' - 'ami-d2423c80',
• 'openfoam171_0-1' - 'ami-28423c7a',
• 'openfoam-dev-1.5' - 'ami-2c423c7e',
• 'openfoam-1.6-ext' - 'ami-3a433d68',

'us-west-1' :

• 'openfoam171_0-1ubuntu2' : 'ami-bb7e2efe',
• 'openfoam171_0-1' - 'ami-8d7e2ec8',
• 'openfoam-dev-1.5' - 'ami-a97e2eec',
• 'openfoam-1.6-ext' - 'ami-5f60301a'

There is a SourceForge project dedicated to the "Cloud Computing for OpenFOAM (R) users" - CloudFlu

And corresponding documentation page

Dear Alexey,

could you tell me something more about "vulashaka" or how to use openfoam with python.
It would be great to get a starting point.

Another question what is the difference between CloudFlu and balloonfoam

thank you in advance.

elvis

We were forced to change the initial "balloon-foam" project name to the new one - "CloudFlu" according to OpenCFD Trademark Guidelines

Hi Ben,
How'd this experiment go? How does OpenFoam run on the Amazon cloud in terms of performance? Can you also run it in parallel?

Thx

Has anyone tried the Cluster Compute instance? I saw some results for the High-CPU Extra Large Compute instance and they weren't that impressive, but the Cluster Compute instance is supposedly tailor made for CFD type computing.

For some reason I can't seem to connect to the amazon machine image posted on openfoam.com. If someone can help me connect to an AMI with OpenFoam, I'd be glad to run a few tests and share the results.

Hi Kevin,
Could you share the sources of your information, please?

From my turn, I read "Benchmarking Amazon EC2 for high-performance scientific computing" written by Edward Walker.
It is not OpenFOAM specific, but what he found out is "approximately 40%–1000% performance degradation in the EC2 High-CPU Medium Instance runs compared to the NCSA (National Center for Supercomputing Applications, USA) runs"
And I have experienced the same situation when I tested my Amazon EC2 front-end for OpenFOAM - cloudFlu

As a consequence, our conclusion for using Amazon EC2 for HPC could be, "use Amazon EC2 if you have no access to the real cluster, only".
At the same time, this investigation by Edward Walker (Oct 2008) were made before Amazon had introduced specific Cluster Computing Instances (July 2010).
So, there is a real hope that things are changed from that time. Unfortunately, I still could see no data or articles which would indicate this shift.

Ye, I will try to prepare specific Amazon EC2 Cluster Computing Instance with OpenFOAM pre-installed.

Best regards,
Alexey

Hi Alexey,

I'm just going by the description by Amazon:

I don't know how Walker was able to test a cluster of High-CPU Medium Instances, as this description says that clusters can only be done with Cluster Compute Instances. But however he did it, I doubt it compares well to the new Cluster Compute instance. Although I am definitely not an expert on any of this stuff.

I did finally did connect to the OpenFoam AMI last night and got it running on a High-CPU instance with solid results. I'm looking to test the Cluster Compute Instance asap.

Hi Kevin,
You see, cluster notion is not defined in hardware terms. According to wikipedia - computer cluster is a group of linked computers, working together closely thus in many respects forming a single computer could be build. You can make a cluster even from a few interconnected PC (know as Beowulf cluster). What enables group of computers become a cluster is low latency interconnection.

I have used this understanding in my cloudFlu implementation and saw the same idea implemented in the widely used common purpose StartCluster - it is really working.

If you look at the Amazon description on account of Cluster Computing Instance, it will be possible to see specification of "I/O Performance: 10 Gigabit Ethernet", which provides comparable performance with the "Infiniband switch" available at 2008 year, claimed by Walker's in his article. So, once we will prepare a Cluster Computing Instance AMI (with OpenFOAM pre-installed), we will be somehow close to the referenced Walker NCSA cluster configuration (except CPU performance; Amazon's should be faster).

I just wasn't sure if you could link instances other than the Cluster Compute Instance (as Amazon's descriptions seems to say that is the only clusterable instance). I suppose there's a way to make it happen, but the fact that the Cluster Compute Instance was specifically built for this purpose gives me confidence.

And like you said, the 10 Gigabit I/O performance is comparable to non-cloud clusters used for CFD type calculations. I did see that that is comparable to some Infiniband Switch specs, but I didn't know which was typical at the time of Walker's report. To hear that he probably had something around 10 Gigabits is very encouraging. I suppose the latency could be problematic, but again, the fact that this instance is build for HPC applications gives me hope.

I don't suppose anyone has an openfoam AMI on US East that I could copy. I'm not familiar enough with openfoam, linux, and amazon AWS to figure out how to do this...it was painful enough just to start running on the EU region.

Hi Kevin,

Is this useful perhaps? http://www.openfoam.com/resources/ec2.php

Hi Gijsbert,

That's actually the tutorial I used to run openfoam on the EU region server, but the best I can do on that server is a single 8-core processor. In order to use the cluster compute instance, I have to be on the US-east region server, so I have to copy the instance from EU to US-east somehow.

Hi Kevin,
Here it is - ami-4fad6a26, Amazon EC2 AMI with the pre-installed OpenFOAM-2.0.0 especially dedicated to Cluster Computing Instances ('cc1.4xlarge' type, in Amazon terms). As it was expected, this image could be run in 'us-east-1' region, only.

Important : It is not enough just to run a number of AMI to get a cluster. Configuration of the cluster itself from the given set of machine instances is the next important step. And this requires a lot of Linux experience and accuracy. I could suggest to use cloudFlu library (especially developed for cloud based HPC OpenFOAM calculations), to automate all these steps and make your way to cloud computing really simple.

Enjoy,
Alexey

Cluster Computing Instance for OF-1.6-ext
 

Hi Alexey and all the others,

my goal is to do cluster computing on amazon using OpenFOAM-1.6-ext. I found this thread and cloudFlu and am now trying to get it working.

Using cloudflu I only get ami-4fad6a26 working, that was mentioned in the last post.

Am I right that the ami's mentioned in an earlier post by Alaxey for earlier Openfoam versions and the extended project version do not work anymore for the new cluster computing instances (like e.g. cc2.8xlarge).

Did anyone already prepare such an AMI or can give me a bit support in building it up by myself.
Thanks in advance!

Best regards,
Matthias

Hi All

Sorry to bump an old thread, but has any further progress been made on developing AMIs for the EC2 cluster compute instances?

I'm particularly interested in testing an AMI for OpenFoam-ext 1.6, ideally in the EU west region.

cloudFlu looks interesting, but we may have QA issues with switching to OF 1.7.

Hello everyone,
Good Evening!

I didnt want to start a new thread because this topic already existed....

Last week I put together a CentOS 6.5 64-bit AMI with the intention of running OpenFOAM (specifically foam-extend-3.1) on Amazon EC2. The
basic setup is as follows:

1. I started with the official CentOS 6.5 64-bit AMI available on the Amazon servers which contains a minimal installation of CentOS 6.5

2. Updated the installation and installed the developer tools, libraries and dependencies required to compile foam-extend-3.1 and the Thirdparty libraries.... everything except Qt and Paraview since my intention was to create a text-based interface.

3. Then, I created a new EBS (Elastic Block Store) Volume and attached it to the Installation in [2]

4. In then split up this volume into two partitions, one which is mounted into "/opt/Simulation" and the other which gets mounted as "/home", in order to separate the User home directories from the base installation.

5. Now, in "/opt/Simulation", I created a folder called "site" in which I created a file called "sim_settings" to act as the main interface to the entire Simulation software installation.

6. In order to separate the whole Simulation setup from the base installation, I created a file in "/etc/profile.d" (which gets called automatically each time a terminal is opened) which simply calls the "sim_settings" file in [5].

7. In "/opt/Simulation", I created a folder "OpenFOAM" under which a new folder "foam-extend-3.1" was created containing a clone of the foam-extend-3.1 git repo. I then compiled foam-extend-3.1 and all the Thirdparty software (except for Qt and Paraview since my intention was to create a text interface).

8. Once all this was done, I unmounted and detached the additional EBS volume created for the simulation setup, and created a new AMI of the base installation of CentOS 6.5 with the modifications required to mount and call the simulation setup (if available... if not, it runs as a normal CentOS installation without the simulation stuff).

9. Finally I made a snapshot of the EBS Volume which now contains a fully compiled and working copy of foam-extend-3.1 and the Thirdparty utilities along with the scripts needed to source them all bundled into a single script called "sim_settings" which is called from "/etc/profile.d".

So basically, I ended up with one AMI (and the corresponding snapshot) of a modified CentOS 6.5 x86_64 installation, and one snapshot of an EBS Volume containing OpenFOAM and its dependencies.

In order to run the whole system, I need to launch an instance of the AMI after attaching the EBS volume snapshot to it.


What do you think of this workflow / concept for running OpenFOAM on Amazon EC2? Do you think this method of splitting the two parts is advantageous and eases maintenance? For example, such an approach could be used for testing, or maintaining multiple simulation setups, etc....

What other ideas have you all thought of and implemented for getting OpenFOAM up and running on Amazon EC2 keeping in mind a modular approach?

I would be interested to get some feedback :-) If anyone is interested in this AMI / EBS Volume, maybe there is some way of sharing it on Amazon (I am a complete newbie with regards to Amazon EC2..... :-)!)

Have a great day ahead!

Philippose

(P.S.... Most of the setup was done using a t1.micro instance type, except for the actual compilation... for which I switched to a c3.2xlarge which is an 8 core system with 15GB RAM using a Spot Instance for 8,5 cents/hour...)

Hello Philippose,

thank you for sharing this information.
I don't really understand what you want to achieve with the modular setup. Can you explain it in more detail? For example, how would such an approach ease maintenance?

Currently, I am also setting up a custom OpenFOAM AMI (still struggling) and I will share my experience here once it is done.

Any other experiences with OpenFOAM on EC2?

Best regards,

Kate

Edit:
@Paul: I got the impression the active development for cloudFlu has ceased some time ago, correct me if I'm wrong.

@all: I think the number one reason I am still struggling with ec2 is, that I am missunderstanding the whole concept:
1) Right now, I am installing OpenFOAM on a publicly available AMI
2) Then I plan to create an image from this running instance. This is then my custome AMI which contains OpenFOAM
3) Suppose I run my custom AMI and make a few simulations. How can I accomplish not to loose my simulation data when I am shutting down my instance? Do I even have to consider this problem since my custom AMI is EBS-backed?
4) Or do I have to create a new image from my running instance before every shutdown in order to not loose my simulation data?

I have many more questions, but the above one troubles me the most.

Hello guys,

after some time using AWS, it is running quite well now. But there is an issue I can't overcome.

Most of the times i am running small models (100,000 to 200,000 cells) on compute-optimized instance types. However these all use hyperthreading. Hence I only use a fraction of the physical processing power.

To summarize it, when I am calculating on a c4.large instance-type, I am only using half the vCPUs I am paying for. What am I missing here?

Best regards,

Kate

Hi Kate,
Try penguin computing on demand, it is cheaper than amazon as you pay only for the time your solver is running, openfoam is already setup and configured and the solver runs on physical cores and not virtual machines, try it and you will love it, if you need any help using it just let me know.
Greetings



Sent from my iPhone using CFD Online Forum mobile app