Lessons Learned in the Migration of GIS from UNIX to a Windows Environment

This paper describes our difficulties in accessing GIS information in a UNIX environment from a MS Windows desktop and our initiatives to resolve them. The information consists of spatial libraries and non-spatial Oracle databases on local and on remote servers. We will describe the current technical architecture which has simplified our operational environment. We will also describe the features of a new GIS user Interface that has significantly facilitated the use of GIS technology by our professional and para-professional staff. Lastly, we will describe the future direction for providing access to GIS technology, which consists of web based applications served over an Intranet using Esri's ARCView Internet Map server.

Introduction

Problems with GIS Application in UNIX Environment:

• - Screens look foreign when compared to Microsoft Windows based applications. Software interface does not have the visual and functional consistency found in standard Windows menus.

• - Unstable and slow access to GIS information It is a complex environment requiring data transfers across networks (UNIX server to Windows desktop).

  We were able to stabilize our environment by replacing the software for accessing the UNIX file system from PCs (Novell's NFS software) with a public domain product called "Samba". This program is running on an interim basis in order to move the GIS libraries from the UNIX server to Novell servers. This change significantly improved performance & reliability.

• - GIS applications take a long time to load Because data is distributed over three networks (Novell, UNIX here and in North Carolina). Queries typically take longer than 30 seconds to respond.

• - Difficult to learn / teach Because X terminal emulation software is not intuitive and as mentioned earlier, the application menu navigation looks foreign to PC users.

• - Aging (Motorola chip based) GIS infrastructure is not suitable for GIS implementation Region wide. We are in the process of expanding the use of this technology from a handful of users to over a thousand potential users.

  Current technological improvements in hardware and operating systems provide a unique opportunity for the standard desktop platform to also support computational intensive applications like GIS.

• - Excessive capital and maintenance costs The cost of an NT workstation is about four times cheaper than a comparable UNIX workstation. Having a single Intel based PC architecture for the desktop and GIS applications also provides a scale of magnitude savings in maintenance and support.

Above Problems Being Addressed by Two Initiatives:

• - Migrate all GIS data holdings and applications to a Microsoft Windows Environment

• - Design and Implement a new Microsoft Windows based GIS User Interface

Considerations in Migration of GIS data holdings to a Microsoft Windows Environment:

Because of the large magnitude of disk space needed to store image data, we looked at different technologies for compressing this data and considered three options:

1. Allowing the Novell Operating System to compress image data which shrinks files in a 4 to 1 ratio but incurs additional CPU time during image processing.

2. Re-sample the images using GRID to unproject the data to geographic NAD83 (originally in UTM) and save the resulting image in JPEG format.

3. Re-sample the images using a commercial product like ERDAS or Esri's Spatial Data Analyst and create a set of images in different resolution scales that can be subsequently selected by an application.

- Migration provided the opportunity to standardize GIS coverages to a common projection. We are creating coverages in unprojected geographic coordinates because the original data sources used various projections (NJ State Plane, NY UTM-18, PR State Plane, EPA Hqts Albers Equal Area). By keeping the coordinates in geographic NAD83 we are able to standardize the input and then project data within the GIS application.

- The process for creating unprojected images requires lots of disk space (about 7 times the size of the files) needed for processing. Also highly CPU intensive since each pixel has to be decoded to Geographic coordinates and then reprojected.

- Novell network pre-assigns many server drives, particularly if client PC uses other Oracle applications. This leaves very limited number of network drives for GIS applications.

- GIS libraries in New York office were also made available to the Edison, New Jersey facility, but data transfers using a high speed (T-1) line was unacceptable. This problem was solved by installing a Novell server and creating a mirror copy of the libraries in the Edison, NJ facility. Since many of our coverages are static, we don't anticipate data synchronization problems.

- When we initially moved the GIS libraries to Novell servers we were unable to access these libraries because we didn't have copies of ArcInfo on an NT server. It became necessary to keep two sets of libraries during this transition... Changes were applied to the libraries on the UNIX server and subsequently copied over to the Novell servers using Samba.

- We used an NT server to create an initial copy of the GIS libraries which we then tried moving to the Novell server, but discovered the File System formats were incompatible. Windows NT server is a complex operating system requiring a system administrator with professional knowledge on this product.

Architectural Components of Intel Based GIS System:

Configurations:

IBM 325 server running Novell Netware 4.11
- 200 Mhz Processor; upgradable to two processors
- 2.25 GB Hard Drive; 16X CD-ROM Drive
- 512 MBytes RAM (4x128 DIMMs)
- 16 GBytes disk space using 5 GByte drives into Core RAID (Redundant Array of Independent Disks) Lightning Array

2. Image server containing Digital Orthophotography, satellite images and other raster data:

IBM 325 server running Novell Netware 4.11
- 200 MHZ Processor; upgradable to two processors
- 2.25 GB Hard Drive; 16X CD-ROM Drive
- 1 GBytes RAM (8x128 DIMMs)
- 120 GBytes of disk space using 9Gbyte drives in Core RAID Lightning Arrays

3. Intranet Map Server configured with ARCView, ARCView Internet Map Server, Map Objects and Map Objects Internet Map Server:

Data General AV4800 server running Windows NT 4.0
- Dual 166 Mhz Pentium Processor
- 2.25 GB Hard Drive; 16X CD-ROM Drive
- 256 MBytes RAM
- 64 MByte VRAM video board interfacing the server console
- 8 GBytes additional disk space using 4 Gbyte drives in Core RAID Lightning Array

4. Oracle server housing the GIS database:

Data General AV4800 server running Windows NT 4.0
- Dual 166 Mhz Pentium Processor
- 2.25 GB Hard Drive; 16X CD-ROM Drive
- 256 MBytes RAM
- 8 GBytes additional disk space using 4Gbyte drives in Core RAID Lightning Array

Topology:

GIS Workstations:

GIS User Interface General Attributes:

• - Includes standard views with minimal selected data as starting point for users (such as Northeast with state boundaries and state names, Caribbean with shorelines and island names, and National with EPA Regions, Canadian, Mexican and Caribbean shorelines).
• - Provides intuitive access to Regional data.
• - Regional libraries are presented in a hierarchy by type of coverage... each coverage has a meaningful name and linked to its metadata record via HTML based Help Screens
• - Is fully linked to Agency, Census and other data via transparent connections to Oracle ( Envirofacts Warehouse; Demographics database; EPA Spatial Libraries; R. GIS database)
• - Modular ... applications, tools and project/application-specific elements can be added in a consistent way.
• - Based in ARCView 3.1 as PC Desktop tool for EPA staff.
• - Optimized for Pentiums but acceptable response on 486s, including portable PCs accessing system remotely.

Required functionality for the Interface is broken down into the following categories:

1. - Project/View Management: Functions include opening/saving projects, views, tables, charts, documentation, etc. These functions are for the most part provided by Arcview without additional programming, but drop-down menus may need to be added to give users access to all general use projects and applications.

2. - Position, Mapextent and Data Selection: Functions include defining the area on the earth to be included in the users view.

   * The mapextend should be determined by the geographical area defined by the user, not the range of data values.

3. - Data Selection and Display: Functions include determing what data will be displayed, showing relevant documentation about the data to allow the user to decide what layers should be shown, how the data will be represented or symbolized, and how legend information will appear. Also includes ability for users to add their own data.

4. - Querying Tools: Functions that allow users to extract information from selected data layer(s) ranging from simple identify functions that generate statistical summaries and charts such as the Population Estimation and Characterization Tool (PECT); buffering analysis; trend analysis; identify facilities meeting a particular criteria; Etc.)

5. - Regional Applications: Functions include ability to select from a library of Regional applications or projects to open (e.g., Environmental Justice Analysis, Enforcement Targeting, etc.).

6. - On-line Help: Functions include Web-browser based help screens for special functions of the Region 2 customization of Arcview and user-friendly metadata for all coverages. (Does not replace basic help already available within Arcview, but supplements.)

7. - Mapping/Layout:
   Users can define a views area of interest by:
   • - Scale (already built into Arcview)
   • - EPA Region
   • - State(s)
   • - City or town
   • - Street address/intersection
   • - Zip Code(s)
   • - Lat/Long
   • - Coverage (already built into Arcview)
   • - Predefined Project Areas (e.g., LI Sound)
   • - HUC Code(s)
   • - Quad(s)
   • - Quarter Quad

Future Direction...

We have adopted Visual Basic as our primary software development environment and we anticipate future GIS applications to be based on this tool set. We will also be exploring techniques to directly access the GIS libraries that are maintained by the originators of the data and eliminate many of the GIS data layers we currently maintain locally.

We are confident the new GIS user interface will be successful because it is being implemented in the context of all our other work. As this application evolves to a web based tool, it will be just another component of the existing desktop. This is in contrast to other information systems projects currently being developed in unfamiliar groupware technology like Lotus Notes that are not gaining user acceptance.

Acknowledgements

The authors also wish to thank Raj Samayam, Robert Simpson, William Hansen and Frank DeMarco for their review and valuable comments.

Robert Eckman
GIS Network Administrator
US Environmental Protection Agency - Region 2
Information Systems Branch
290 Broadway
New York, NY 10007
Telephone: (212) 637-3324
Fax: (212) 637-3354
E-mail: eckman.robert@epamail.epa.gov