Re: [AMBER] cloud vs purchase

From: Gustaf Olsson <gustaf.olsson.lnu.se>
Date: Thu, 5 Sep 2019 09:16:01 +0000

I suppose all of this depends on how you plan to utilise the resources. Are you planning on doing CPU or GPU calculations, are you planning on running long consecutive simulations or are we talking about shorter but more intensive bursts? It also depends on what kind of cloud provider you are talking about.

I have some experience running evaluations and simulations on a CSCloud installation of Ubuntu. On the system I’m using, as the graphics card is being emulated and therefor there really is no GPU to utilise for calculations and of course I have had some issues with graphics forwarding. Basically I can get X-window to forward for quick checks using grace however trying to use VMD on the VPS was pretty much impossible. If you are planning on running calculations using GPU then make sure there are physical GPUs available for this and not only emulated, I also expect that using emulated GPUs would lead to problems compiling and serious performance limitations.

Not being a “cloud provider”, a computational center I have access to has cluster with some serious GPU power available and in that case software is preinstalled for scientific use. Running GPU calculations here is fast and works smoothly, so it is not impossible as long as the service you select supports the use you are planning.

As graphics cards are emulated on the cloud service I’ve had issues with "visual inspection", like trying to run VMD which has turned out to be basically impossible. This means I have to mount a remote volume and use my local install or downloading the files. What matters here is checking egress/transfer fees and caps, some providers charge extra for traffic (up/down) which might become expensive in the long run if systems are big.

Just for running CPU simulations and cpptraj evaluation works well. In particular using CPPTRAJ OpenMPI for rdf and hbond is a lot faster then using another server center I have access to. I think this has to do with read/write speeds being a bottleneck on that server.

I have no idea about general pricing for cloud computing as the cloud installation I am using for the VPS belongs to my university though looking at DigitalOcean, running a machine using the specs:

MEMORY VCPUS SSD DISK TRANSFER PRICE
96 GB 20 vCPUs 1920 GB 10 TB 480/month

This would cost $ 5760 for per year, so if you are planning on utilising resources if this caliber then I would make a one time investment of $6000 which should pay for itself within a little more than a year. Still, I have no idea what places are charging for the resources you would like to use so I can’t really advice you there. On the plus side, hardware maintenance is not you problem. Things can still do wrong even with a cloud provider though I suppose that this is something you could avoid.

If you are working for a university and not trying to do this privately or at a company, then do some research. There are “publicly” available computational resources available to researchers in a lot of places where you can simply apply for resources.

Not sure if this helps you at all but it is at least a little feedback
// Gustaf


> On 5 Sep 2019, at 09:25, Fabian Glaser <fabian.glaser.gmail.com> wrote:
>
> Dear Amber experts,
>
> I would like you about your experience about running amber extensive simulations on a cloud (amazon or any other) vs purchasing a new GPU x 4 machine. I have no experience at all with clouds.
>
> Does anybody uses clouds for extensive MD?
> Do you have a cost estimation comparison (per ns x day for example)?
> How difficult it is to deal with implementation and compilation issues on a cloud?
>
> My comparison should be based on purchasing a new GPU X 4 machine, for which I would invest let’s say about 6000$, and comparing it to running on a cloud instead.
>
> Thanks a lot!!
>
> Best,
>
> Fabian
>
> Fabian Glaser PhD
>
> Bioinformatics Knowledge Unit - BKU
> The Lorry I. Lokey Center for Life Sciences and Engineering
> Technion - Israel Institute of Technology, Haifa, Israel
> Web http://bku.technion.ac.il/
> Tel +972 (0) 4 8293701
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber

_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Sep 05 2019 - 02:30:01 PDT
Custom Search