U. Herbert Grote and Ronald Kahn(1)
National Oceanic and Atmospheric Administration
Forecast Systems Laboratory
Boulder, Colorado
Table of Contents
Part of the mission of the Forecast Systems Laboratory (FSL) is to transfer proven forecasting techniques and technologies from the meteorological research environment to operations. FSL currently has two systems in use at National Weather Service forecast offices for evaluation, one at the Denver, CO NWSFO and the second at the Norman, OK NWSFO. The systems were installed approximately five years ago as part of the AWIPS risk reduction activities. Each of the systems consist of over a dozen DEC microVAX computers and several Ramtek display devices. These systems are reaching the end of their life spans. The networks are inadequate for transferring significant additional data and the processors do not have sufficient computing power to provide the additional processing required to add major new capabilities. Also, since this hardware and software technology is very different from AWIPS it is difficult to transfer software and experiences to AWIPS. For these and other reasons it was necessary to create a new testbed for evaluating these capabilities.
The goal of this new development is to provide a powerful forecaster workstation that provides a complete set of functionality for operational weather forecasters. The system will have most of the capabilities of the AWIPS system and will exceed the AWIPS capabilities in some areas. In order to have maximum input into the AWIPS development effort the workstation must be developed in a very short time frame. FSL conducted a forecast exercise using the new workstation, known as WFO-Advanced (MacDonald and Wakefield, 1996), last October and will install a workstation in the Denver Weather Forecast Office in early 1996.
The following features are deemed necessary to create a workstation with a reasonably complete set of functions for operational forecasters.
The WFO-Advanced will be able to acquire conventional and experimental data on the national and local scales. The primary national datasets include observations and guidance products obtained from the AWIPS Satellite Broadcast Network (SBN). Local datasets include the WSR-88D radar, mesonets and hydrological observing systems. Among the experimental datasets under consideration are gridded data from the Local Analysis and Prediction System (LAPS), and the Mesoscale Analysis and Prediction System (MAPS), Continental United States (CONUS) radar products, and moisture fields derived from the Global Positioning System (GPS) receivers.
The workstation will provide the capabilities needed by operational forecasters to perform their daily tasks. The forecaster must be able to access all meteorological data, manipulate them to extract the desired information and display them on an integrated display. Forecaster often need to execute custom application programs to extract the desired information. Once all essential information has been gathered and reviewed, the workstation must provide tools to assist the forecasters in their preparation of forecasts, watches and warnings. Many of these products will be prepared on a text processor and disseminated to other forecast offices and end users.
The WFO-Advanced provides a set of development tools, such as compilers, debuggers and software libraries, that allows expansion of the basic set of diagnostic and forecasting utilities. The application programs must be able to access the real-time database and they must integrate easily with the workstation environment. Safeguards will be incorporated to reduce the chance of applications crashing the system or tying up valuable computer resources.
The workstation will also have the capability to create forecast products through graphical interaction with a gridded database. Computer support for forecast preparation, as is done in the AWIPS Forecast Preparation System (AFPS), will include generation of service-specific weather elements from MOS and model grids (Wier and Wakefield, 1996), graphical editing of forecasts, generation of product parameters and the creation of text products.
FSL's local analysis and prediction system (Schultz, 1996) will provide high spatial (10 km or less) and high temporal resolution gridded dataset over a local domain (Snook, 1996).
The workstation will provide the capability to display gridded data fields in three dimensions. The forecaster will be able to interact with the three dimensional fields and perform such functions as rotation, translation, zoom, sampling and also initiate execution of special diagnostic tools.
Last, the workstation will be able to routinely disseminate gridded data fields from the local analysis and prediction analysis or the forecast preparation components. FSL has developed a PC-based display system (Subramaniam and Jesuroga, 1995) for use by emergency managers that can display these grids and other related data.
Development of the system requires that a basic system architecture be defined first. Some of the fundamental questions to be answered concern the number of processors needed, the type of network required to connect these processors, where the database should reside and what display controller architecture provides the desired performance?
In order to provide a comprehensive set of functions in a short time and with limited resources, as much off-the-shelf software as possible must be incorporated. Excellent meteorological software packages have been developed within NOAA and other organizations that can be integrated into the workstation. Use of loosely coupled integration, such as sharing only the database and display device by software packages, provides an expedient and often acceptable way of software integration.
Off-the-shelf and available components include both, computer hardware and software. They include:
SBN and WSR-88D subsystems, which are being developed by the AWIPS contractor, PRC. The SBN consists of the ground station, communications board, and an HP743i acquisition processor with associated software. The WSR-88D radar subsystem consisting of an HP743i, communications board, and acquisition software.
Available software packages include GEMPAK (developed by NASA and NMC), the Shared Window Server (developed by the NWS' Advanced Demonstration and Development Laboratory), a hydrological applications package (developed by the Office of Hydrology), the LAPS, the AFPS and the PC-based dissemination system (all developed at FSL).
The following workstation components are new: local and experimental data acquisition, 2-D data display and interaction, 3-D data display and manipulation, text editing, and interactive applications.
Resource requirements (i.e. processing, display, memory, and storage) for these components can be determined from actual measurements, based on requirements for similar functions, or system specifications, such as data rates and retention times. For some components, such as models, a suitable first-guess estimate can be obtained by extrapolation from existing processing demands.
Processing - Table 1 provides an estimate of the processing requirements for the workstation. The greatest anticipated processing requirement is for the local modeling and the display generation.
Display - The display resource requirements for the workstation are significant. The workstation must be able to display images in multiple X-windows with the proper colors, even when a window is not active. True color systems readily meet this requirement. However, performance of true color systems is far less than that of pseudo color systems for many simple operations such as changing the image color. To meet the performance requirement the WFO-Advanced workstation will use true color only for four small windows (200 by 200 pixels). A large (920 by 920 pixels) window will use two 8-bit color look-up tables (underlay and overlay).
In addition, the workstation must concurrently satisfy the display requirements of existing software packages. The
Table 1. Processing Requirements
previously identified existing software packages have a minimum requirement of 8-bit pseudo color.
The 3-D components require a display controller with 24 bit planes of refresh memory and a graphics drawing capability of approximately 2 million vectors per second. Although 3-D visualization is possible with a slower system, interactivity suffers accordingly.
Memory - The largest amount of memory is needed for data caching for the 2-D display component. In order to provide the necessary performance for zoom and progressive disclosure of data all displayed data and associated animation frames must be cached in memory. The minimum cache size, based on a mix of products, is 135 megabytes. An additional 64 megabytes is required for the operating system and 2-D processes.
Storage - To routinely save the last two days' worth of data the workstation's database must be able to store three days' worth of data. Table 2 provides an estimate of the storage requirements for a single day. These estimates are based upon the data being stored in netCDF format.
The database is the heart of the workstation. All real-time data must flow to the database and all data requests from the workstation components must be served by the database. The architecture and performance of the database is critical to successful operation of the workstation. Consequently, to simplify data management all data will be stored in a central repository. A major concern is reduced data retrieval time as a result of contention for the disks and network communications port. Tests on an HP 9000/755 indicate that the data transfer rate may drop by a factor of three if two simultaneous Network File Server requests are made to a processor.
Table 2. Database Sizing
Figure 1 provides a performance comparison of various workstations from HP and other vendor's based on each vendors rating of their machine(s).
Figure 1. Performance Comparison of Various Workstations
(Click on figure to see a legible version)
SPECfp92 and SPECint92 ratings provide only a rough comparison of workstation performance. Other factors, including memory capacity, I/O performance, graphics drawing speed and operating system capabilities need to be considered carefully before making a final selection.
For large data storage the alternatives include disk arrays with or without RAID controllers, independent disks, and various vendors' special purpose solutions.
The number of processors in the system is based on the number of workstation components, compatibility of components, execution frequency, throughput requirements, cost and reliability considerations. For example, compatibility between data acquisition and display components is low. Data acquisition requires the buffering and processing of real-time events while display processing is driven by actions initiated by the forecaster. Many of the components, such as 2-D display, 3-D display, 3-D editing, graphical forecast preparation, and GEMPAK do not need to execute concurrently since it is unlikely that the forecaster will perform all these functions at the same time.
Several trade-offs exist with regards to the database. The database can be centralized or distributed, some data may be routinely transmitted to several local storage devices upon arrival to improve load performance, and data may be stored in different ways on the disks. In order to reduce disk and communications contention on the data server the workstation must routinely distribute certain high volume data to local storage devices. Loading some of the data into local memory cache will further reduce retrieval times.
Figure 2 illustrates three different approaches to integrating the database for independently developed applications (components).
Figure 2. Database Access Strategies
(Click on figure to see a legible version)
One approach is to allow each application to have its own data interface and separate database. A second approach is to allow each application to have its own interface and share one common database. The third approach is to provide a common interface to the applications and one (or more) common database. The second option was chosen since it minimized storage requirements and will have minimum impact on the existing applications. In making the various hardware and software trade-offs it is necessary to consider the impact of each decision on AWIPS compatibility.
Figure 3 illustrates a proposed hardware architecture for a Weather Forecast Office.
Figure 3. WFO-Advanced System Architecture
(Click on figure to see a legible version)
The acquisition/CP processors are HP743i computers as specified for AWIPS. An HP 725 is used for experimental and local data acquisition and the meteorological applications processors. Because of better I/O performance an HP 755 is used for the data server. Two HP725 computers and an X-terminal comprise a single user station. Because performance was deemed critical each display was assigned a dedicated processor. The acquisition processor and X-terminal are connected by an Ethernet, all other processors use the Fiber Distributed Data Interconnect (FDDI). A FDDI concentrator with an Ethernet bridge router provides the basic interconnect mechanism.
To achieve the high reliability needed for a forecast office redundant processors are specified for the data server and applications processors and a spare processor is provided for data acquisition.
The software development approach most nearly follows that of the Spiral Model (Boehm, 1988), which includes iterative prototyping, a process extremely valuable in ensuring that the system is usable and meets user needs. Many user interface builders, such as tcl/tk used by FSL, exist that assist with prototyping the user interface.
The HP-VUE utility can be used to integrate independent software packages at the display. This software creates independent workspaces for each of the software packages that can be displayed or hidden from view with a simple mouse selection. A key element in integrating software packages with HP-VUE is to ensure that sufficient color resources exist to provide smooth transitions between workspaces.
To allow uninterrupted software development, integration and quasi-operational testing, three separate environments are required (Williams and Davis, 1996). Configuration control and reliability increase as the software progresses from development to quasi-operational testing.
The WFO-Advanced system provides a powerful new test platform for testing promising new forecasting techniques. The system is similar to AWIPS and integrates specific AWIPS hardware and software components. This similarity promotes technology transfer between FSL and AWIPS. The system configuration to be tested in at the Denver WFO in the early part of 1996 may vary slightly from that defined here. The proposed hardware configuration for AWIPS will include the new HP J200 and K200 processors. A J200 workstation processor would drive two graphic display devices. It is expected that the WFO-Advanced software will support this type of configuration.
Boehm, B. W. 1988: A Spiral Model of Software Development and Enhancement, IEEE Proceedings, 61-72.
MacDonald, A. and J. Wakefield, 1996: The WFO-Advanced: An AWIPS-like Prototype Forecaster Workstation. Preprints, 12th International Conference on Interactive Information and Processing Systems for Meteorology, Oceanography and Hydrology, January 28 - February 2, 1996, Atlanta, GA, Amer. Meteor. Soc.
Schultz, P., 1996: Local Data Analysis and the Mesoscale Model on the WFO-Advanced Workstation. Preprints, 12th International Conference on Interactive Information and Processing Systems for Meteorology, Oceanography and Hydrology, January 28 - February 2, 1996, Atlanta, GA, Amer. Meteor. Soc.
Subramaniam, C. and R.T. Jesuroga, 1995: The Dissemination Project: A Decision Support Tool for Emergency Managers. Proceedings, Fifth Topical Meeting on Emergency Preparedness and Response, April 18 - 21, 1995, Savannah, GA, Amer. Nuc. Soc., 407-411.
Snook, J.S., 1996: Local Domain Forecasting Support to the 1996 Atlanta Olympic Games. Preprints, 12th International Conference on Interactive Information and Processing Systems for Meteorology, Oceanography and Hydrology, January 28 - February 2, 1996, Atlanta, GA, Amer. Meteor. Soc.
Wier, S. and J. Wakefield, 1996: Using Numerical Model Output to Provide Initial Forecasts of Surface Weather for the AFPS. Preprints, 12th International Conference on Interactive Information and Processing Systems for Meteorology, Oceanography and Hydrology, January 28 - February 2, 1996, Atlanta, GA, Amer. Meteor. Soc.
Williams, S., and D. Davis, 1996: The Development of the WFO-Advanced Hydrometeorological Workstation. Preprints, 12th International Conference on Interactive Information and Processing Systems for Meteorology, Oceanography and Hydrology, January 28 - February 2, 1996, Atlanta, GA, Amer. Meteor. Soc.
Footnotes
- (1) Contract with Science and Technology Corporation, Hampton, VA 23666
This document is maintained by Joe Wakefield.
Last updated 15 Feb 96
