Yes, this has been done, but due to the way network calculations currently work, it doesn't provide as big a speedup as you might hope.
Here's a summary of things written about it on the UF mailing list earlier.
Samuel Monnier (March 2, 2014):
I don't know if this is well-known, but here is a video explaining how
to use Amazon EC2 to perform network computations with Ultrafractal: https://www.youtube.com/watch?v=le36JoO2dgs
The interface changed a bit, and you'll have to create your own "AMI", i.e. system image with Ultrafractal installed, because the one
mentioned in the video apparently disappeared.
It works well fractals such that each pixel takes a very long time to
render, typically deep zooms. I doesn't work well at all for fractals
whose pixels render fast. For instance, if you want to compute a huge image out of such a fractal, or if you are rendering an animation, the network feature of UF5 is of little use. This is because UF5 sends very small groups of pixels to be computed to the remote computer at a time, so if these are computed quickly, the network delay slows down computations considerably. I wish there was a feature in UF5 allowing to adjust that, because it's only a matter of the size of the pixel packets you send to the remote computer. (In case Fredrick is listening...)
For instance, I tried to use a c3.2xlarge computing instance, which is
supposed to be 10 times faster than my laptop. Using it to render an animation I could only roughly double the computation speed. I found that the optimal number of connections in UF5 is in this case between 9 and 15. Less or more connections effectively increase the computing time.
Damien Jones wrote:
Having used Amazon EC2 for things other than fractals, I'll just say
that if you decide to do this, you should probably make the AMI (like a "snapshot" of the computer you want to use) yourself, starting from a reliable, clean setup. This will do two things:
You will know nobody has tampered with the Windows setup, such as installing backdoors. (If you don't care about your parameters being copied, this isn't as much of an issue.)
You will know UF is set up using your registration key, rather than someone else's.
Also, be sure you read up on the cost of running the server, and make SURE you shut down the EC2 server when you are not using it... otherwise you will get a bill at the end of the month you'll be unhappy with. Some of the options you can set on the EC2 servers are very expensive.
Samuel replied:
- You will know UF is set up using your registration key, rather than someone else's.
That's not really an issue, because you need only to run the server
version of UF, which does not require registration.
Also, be sure you read up on the cost of running the server, and make SURE you shut down the EC2 server when you are not using it... otherwise you will get a bill at the end of the month you'll be unhappy with. Some of the options you can set on the EC2 servers are very expensive.
That's good advice indeed and is mentioned in the video. One should also be careful to "terminate" the instance, instead of "stopping" it. In the latter case, an image of the system is kept, which incurs extra costs (athough much smaller ones than leaving an instance running).
Hopefully this helps!
Yes, this has been done, but due to the way network calculations currently work, it doesn't provide as big a speedup as you might hope.
Here's a summary of things written about it on the UF mailing list earlier.
Samuel Monnier (March 2, 2014):
> I don't know if this is well-known, but here is a video explaining how
to use Amazon EC2 to perform network computations with Ultrafractal: https://www.youtube.com/watch?v=le36JoO2dgs
The interface changed a bit, and you'll have to create your own "AMI", i.e. system image with Ultrafractal installed, because the one
mentioned in the video apparently disappeared.
> It works well fractals such that each pixel takes a very long time to
render, typically deep zooms. I doesn't work well at all for fractals
whose pixels render fast. For instance, if you want to compute a huge image out of such a fractal, or if you are rendering an animation, the network feature of UF5 is of little use. This is because UF5 sends very small groups of pixels to be computed to the remote computer at a time, so if these are computed quickly, the network delay slows down computations considerably. I wish there was a feature in UF5 allowing to adjust that, because it's only a matter of the size of the pixel packets you send to the remote computer. (In case Fredrick is listening...)
> For instance, I tried to use a c3.2xlarge computing instance, which is
supposed to be 10 times faster than my laptop. Using it to render an animation I could only roughly double the computation speed. I found that the optimal number of connections in UF5 is in this case between 9 and 15. Less or more connections effectively increase the computing time.
Damien Jones wrote:
> Having used Amazon EC2 for things other than fractals, I'll just say
that if you decide to do this, you should probably make the AMI (like a "snapshot" of the computer you want to use) yourself, starting from a reliable, clean setup. This will do two things:
> 1. You will know nobody has tampered with the Windows setup, such as installing backdoors. (If you don't care about your parameters being copied, this isn't as much of an issue.)
> 2. You will know UF is set up using your registration key, rather than someone else's.
> Also, be sure you read up on the cost of running the server, and make SURE you shut down the EC2 server when you are not using it... otherwise you will get a bill at the end of the month you'll be unhappy with. Some of the options you can set on the EC2 servers are very expensive.
Samuel replied:
> > 2. You will know UF is set up using your registration key, rather than someone else's.
> That's not really an issue, because you need only to run the server
version of UF, which does not require registration.
> > Also, be sure you read up on the cost of running the server, and make SURE you shut down the EC2 server when you are not using it... otherwise you will get a bill at the end of the month you'll be unhappy with. Some of the options you can set on the EC2 servers are very expensive.
> That's good advice indeed and is mentioned in the video. One should also be careful to "terminate" the instance, instead of "stopping" it. In the latter case, an image of the system is kept, which incurs extra costs (athough much smaller ones than leaving an instance running).
Hopefully this helps!
