What is Cloudrun?
Cloudrun is the cloud computing platform for weather and ocean prediction models. Our goal is to help our users overcome two barriers in running prediction models:
- The need for a supercomputer to crunch the numbers
- The need for human expertise to configure and run the models
Both are extremely scarce and expensive. First, we remove the need for a dedicated supercomputer by spinning up high performance compute nodes in the cloud and using them only for the duration of the run. Second, the configuration and running of the model is automated and baked into the Cloudrun platform itself.
What models do you currently support?
We currently support the Weather Research and Forecasting (WRF) model versions 3.8.1, 3.9, and 3.9.1. We will include support for new versions as they are released by the WRF development team at NCAR.
Do you plan to support other models?
Yes, we plan to offer other models, like ocean, wave, and climate models, in the future. If you would like to be able to run other models with Cloudrun, please let us know at firstname.lastname@example.org.
What is WRF?
Weather Research and Forecasting (WRF) model is an advanced numerical weather prediction and modeling software suite. It features a non-hydrostatic dynamical solver and a wide array of boundary layer, cloud microphysics, and cumulus parameterization schemes. WRF can run in both global and regional configurations, as well as with storm-following nests for high resolution tropical cyclone predictions. WRF has been developed by the National Center for Atmospheric Research (NCAR), with significant contributions from the atmospheric research community. It is a registered trademark of the University Corporation for Atmospheric Research.
What dynamical cores of WRF do you support?
Currently, we support only the Advanced Research WRF (ARW) core, as it is by far the most widely used mode that WRF is run in. If you would like to run WRF-NMM or HWRF through Cloudrun, let us know!
What physics and dynamics options can I use through Cloudrun?
You can use any physics and dynamics options that you would otherwise use when running WRF yourself. Our system will attempt to catch any errors or incompatibilities that you may have in your
Can I run fixed or moving nests?
Yes! You can run enable any number of fixed or moving nests that WRF itself would allow you.
Can I run WPS (WRF Preprocessing System) with Cloudrun?
WPS capability with Cloudrun is currently under development.
Can I upload my own, custom version of WRF?
Currently, you can only run the official, unmodified WRF releases. If there's enough demand, we'd love to give users the ability to run their custom WRF versions. Please let us know at email@example.com if this capability is important to you.
What is the difference between different run times and costs?
In most cases, we will offer you a few different options of run times and prices. Runs that take shorter time to complete run on larger number of parallel processors, and thus cost more. Actual time to complete may vary somewhat from the estimated time, however you will not be charged extra if the run takes longer to complete. You will always be charged only the price that you select before starting the run.
How do you determine the price of each run?
When you upload your configuration file (
namelist.input for WRF) our system parses is and calculates estimated compute times for several different configurations of parallel processors. The price that you pay includes the bare compute cost of the servers, and the cost of Cloudrun as a product.
I noticed that the price for a run has changed from before, even though the run has the same size. Why?
Compute times depend on many different factors, both in terms of model configuration, and the network and infrastructure performance on our end. If the price does not look correct for any reason, please let us know at firstname.lastname@example.org and we will look into it.
Will I get charged if my run ends prematurely?
If you choose to stop your run before it completes, you will be charged only for the fraction up to that point, pro-rated to the second. However if the run crashes due to software or hardware failure on our end, you will not be charged for that run.
Do you offer monthly or annual plans?
We currently offer a pay-as-you-go plan. Under this plan, there is no monthly fee. You pay individual charges for each run, and for storage of model output. We may offer monthly and annual plans in the future.
Will you have a public API for users and developers?
Yes! We'll be opening our API for our users to create and access model results directly. If you'd like to be involved in our private beta API as soon as its available, please contact us at email@example.com
What programming languages will your API support?
Our API will support HTTP requests (for example with
curl) and we have a Python interface in development. Send as a request if you'd like your favorite language to be supported.
On what kind of system do you run models on?
All model runs are done on a custom minimal Linux system using the latest generation Intel Xeon Platinum 8168 processors, released in Q3 2017.
Do you use virtualized or bare-metal servers?
For model runs, we use virtualized servers to minimize overhead and cost of the service and maximize scalability. However, our instances are always dedicated and run at near 100% CPU usage (no CPU throttling).
Do you use parallel computing for your runs?
Yes, all model runs are done in parallel.
How does your hardware compare with industry-standard supercomputers?
Our processors are latest generation and typically few years ahead of industry-standard supercomputers. This is because classic HPC systems are very large and can update hardware only every 3-5 years or so. For example, DOE's recent system Cori uses Haswell processors from 2014. NASA's largest and fastest supercomputer Pleiades is mostly based on Ivy Bridge processors from 2013, with some Haswell and Broadwell processors. While Cloudrun does not aim to be a replacement for those systems, we do aim to provide much faster and cheaper solution using state-of-the-art hardware and the flexibility of the cloud.
Do you offer customer support?
We currently offer free e-mail support to all Cloudrun users. You can also reach us by phone at +1 (760) 853-1067, however we do not yet offer dedicated phone support and may not answer every call.
Do you offer scientific support and consulting?
We currently do not offer scientific support or consulting. We plan to offer dedicated scientific support as part of premium membership plans.
The Cloudrun Team
Who is Cloudrun?
Cloudrun was founded by Milan Curcic and Josh Noe. Milan is a meteorologist and oceanographer with expertise in High Performance Computing (HPC) and Earth system model development. Josh is a computer scientist with expertise in designing and building complex full-stack systems and user interfaces.
How to best reach us?
If you need help with using Cloudrun or experience an issue with the platform, the best way to reach us is by email at firstname.lastname@example.org, or by phone at +1 (760) 853-1067. Keep in mind that we currently don't have a dedicated phone support. If we can't pick up, please leave us a message and we will get back to you as soon as we can.