|
| 1 | +************************** |
| 2 | +Frequently Asked Questions |
| 3 | +************************** |
| 4 | + |
| 5 | + .. dropdown:: How much does it cost? |
| 6 | + |
| 7 | + The main costs include hosting a web domain, compute cluster time, and data egress. A web domain typically costs $12/year. |
| 8 | + The forecasts are run on AWS's hpc6a clusters that cost $3.00/hour. The data egress costs $0.09/GB. The total cost depends on user interaction with the forecast viewer. For example, to view every variable that is currently available for a 24-hour forecast with hourly output, the system uses about 700MB, which translates to about $0.06. |
| 9 | + |
| 10 | + To give users a general idea of the range of costs, here are some example forecasts and their associated costs to compute. Note that these estimates are for running the forecast, which takes place on the most expensive part of the system: the compute nodes. Additional costs for tasks like viewing products are generally minimal and are described above. |
| 11 | + |
| 12 | + +----------+---+-------+-----+----+-------+-----+------+ |
| 13 | + | Name |Res|Grid |Fcst |Out |Wall |Cores|Cost | |
| 14 | + | | |Size |Len |Freq|Clock | | | |
| 15 | + +==========+===+=======+=====+====+=======+=====+======+ |
| 16 | + | Caribbean|6km|700x500|24 hr|1 hr|2.17 hr|96 |$6.91 | |
| 17 | + +----------+---+-------+-----+----+-------+-----+------+ |
| 18 | + | Brail |3km|500x520|24 hr|1 hr|1.62 hr|96 |$5.11 | |
| 19 | + +----------+---+-------+-----+----+-------+-----+------+ |
| 20 | + | Denver |1km|250x250|24 hr|1 hr|1.53 hr|48 |$4.71 | |
| 21 | + +----------+---+-------+-----+----+-------+-----+------+ |
| 22 | + |
| 23 | + .. dropdown:: Can you run multiple domains? |
| 24 | + |
| 25 | + Currently we support running single domains only. Multiple domains are on the list for future enhancements. |
| 26 | + |
| 27 | + .. dropdown:: What is the input data used? |
| 28 | + |
| 29 | + Initial conditions are provided by the Global Forecasting System (GFS) at 3-hourly time intervals. Future enhancements may allows for 1-hourly interval data to be used. |
| 30 | + |
| 31 | + .. dropdown:: Can the input data source be changed? |
| 32 | + |
| 33 | + No, at this time the system only uses the Global Forecasting System (GFS) for initial conditions. But future enhancements could allow for varying sources. |
| 34 | + |
| 35 | + .. dropdown:: Can you download the model output? |
| 36 | + |
| 37 | + At this time, the model data is not saved. Only the forecast plots viewable on the Forecast Viewer are saved. But, we expect to include the option to save model output in the next released version. |
| 38 | + |
| 39 | + .. dropdown:: What cloud service providers are supported? |
| 40 | + |
| 41 | + Currently the system only works on Amazon Web Services (AWS). Interested in porting it to another CSP? We are too! Let’s talk, send us an email. |
| 42 | + |
| 43 | + .. dropdown:: What model is used? |
| 44 | + |
| 45 | + The system is currently designed to use the WRF model. |
| 46 | + |
| 47 | + .. dropdown:: A WRF job failed, how do I fix it? |
| 48 | + |
| 49 | + WRF jobs can fail for many reasons. First, look at the log files for the failed job on the WRF Jobs page. The status of the job may help indicate which component has failed, e.g. metgrid, real, or wrf. In the :ref:`ui_log_viewer`, search the log files for that component first for a clue as to what went wrong. If the job is based off a new configuration (model domain, physics, etc.), it's possible that those settings aren't properly tuned. There are countless ways to configure the model. Properly tuning a new configuration requires some knowledge of WRF and NWP modeling. Refer to the `WRF documentation <https://www2.mmm.ucar.edu/wrf/users/docs/user_guide_v4/v4.4/contents.html>`_ for help. |
| 50 | + |
| 51 | + Common errors you may encounter include: |
| 52 | + |
| 53 | + * **Over-decomposition:** The error *"domain size is too small for this many processors"* or something similar may appear in the *Real* rsl.* files when the *Real* process fails. This means that too many processors were used for the domain size. If the *Set automatically* checkbox was selected for the Core Count in the model configuration GUI, it is possible that it overestimated. :ref:`Modify the configuration<ui_update_existing_config>` core count to uncheck *Set automatically*, manually set it to a smaller value, save, then :ref:`run a new forecast<run_wrf>` with the updated configuration. |
| 54 | + * **CFS errors:** "cfl" errors may occur in the *WRF* rsl.* log files. This means the model became unstable which can happen for a lot of reasons and often requires advanced debugging. The `WRF Forum <https://forum.mmm.ucar.edu/>`_ may have some clues. |
0 commit comments