GEOG 865
Cloud and Server GIS

Ways to serve maps and the role of tiled services


By this point in the course, you may have observed that there's more than one way to take raw GIS data from your server and put it into a map in someone's web browser. Recall some of the map services you used in the previous two lessons:

  • The AppalachianTrail and BighornReferenceLayers services used the server to draw a new map every time the user panned or zoomed. The completed map image was then sent to the user's computer for immediate display in the web browser.

    In ArcGIS Server-speak, this is a dynamic map service, because it is drawn "dynamically", or on-demand, by the server. Dynamic map services are the default that you get when you publish a map service to ArcGIS Server, but it's not just Esri services that use this pattern. The Open Geospatial Consortium (OGC) Web Map Service (WMS) also uses this approach of sending back a dynamically-drawn image based on URL parameters sent from the requesting computer.

    Dynamic map services are easy to set up, but they are not the fastest or most scalable type of service (scalability refers to how many clients can be served at one time). Accessing data and drawing a map image require the server to do work, and if a large number of people are requesting maps at the same time, the server can get overwhelmed. Complex maps with many layers or fancy symbols may take the server an unreasonably long time to draw.

    Recognizing that dynamic drawing speed was suboptimal in early versions of ArcGIS Server, Esri wrote a streamlined drawing engine for the server and introduced it as an option at 9.3.1 using a special type of file you had to explicitly create called an MSD (map service definition). Starting from 10.1 the optimized drawing engine is used in all map services and the MSD takes a behind-the-scenes role. However, even a dynamic service, based on the optimized drawing engine, may not be appropriate for some heavily-used or complex maps.
  • The BighornHabitat layer from the previous lesson containing sheep sightings and habitat was not drawn dynamically by the server. Instead the coordinates of all the vertices of the points and polygons were sent to the web browser as text, and the browser did the work of drawing the features. This approach can be thought of as using client-side graphics, because the client (web browser) uses its graphic-drawing technology to put the features on top of the map.

    ArcGIS Server map services can optionally be drawn as client-side graphics. OGC Web Feature Services (WFS) are also designed to be queried and drawn by the client machine.

    Client-side graphics are great for providing interactivity to your web applications. Once you get all the geometry and attributes of the features onto the client machine, you don't have to go back to the server any more. You can change the shape or color of a feature in real time, allowing for highlighting of features, reclassification, or making edit sketches.

    Client-side graphics are not appropriate in every case. Most browsers have a limit on how many points or vertices they can draw before they really start to slow down. Also, you saw in the previous lesson how symbology was limited using client-side graphics. The diagonal hash line symbolizing the habitats could only be drawn with a simple solid fill. This is because web browsers know how to draw a limited set of symbols; for example, they can't use the whole gamut of ArcMap polygon fills.
  • The imagery service from ArcGIS Online that you used as a background for your bighorn sheep map was a tiled (or "cached") map service. In these types of services, the server just hands out little square images (tiles) of the map that it has stored in a cache on disk. These images are usually generated by the server administrator at a number of different scales before the service is made available to the public. Then when web users request to see a map, the server does not have to do the work of drawing the map; it can just send back whichever tiles are needed to fill the map request.

    A server can send out tiles a lot faster than it can draw maps; therefore, tiled map services are quicker and can accommodate a lot more users than the other two types of services. Just take a look at how fast the imagery layer loads in your bighorn sheep app, or how quickly all the basemaps appear in the map viewer. If you have a complex map (especially a basemap) or a map that will be viewed by a lot of people, it's usually worth the effort to make a tile cache.

Like the two other choices above, tiled maps also have their unique drawbacks. The biggest one is the time investment and server power needed to generate the cache, along with the disk space necessary to store it. Also, because a cache represents a snapshot of your data at one point in time, it requires maintenance. If your source data or your map symbology is edited, you have to update the corresponding tiles in order for people to see the changes.

In this lesson you'll learn about designing a map with the goal of building a tile cache. You'll get a chance to make some tiles and use them on the web. Since the number of tiles in a cache can multiply with each scale level added and become unmanageably large, you'll also learn about strategies for building and updating very big caches.

A word about vector tiles and rasterized tiles

A word about different tile types before we begin: There are two main types of tiles commonly used in web maps today. The kinds of tiles we've been talking about above can be thought of as rasterized tiles; in other words, they are images made up of grids of pixels. Rasterized tiles are easy for clients to draw because most apps and all web browsers know how to display an image like a JPG or a PNG; however, the server has to construct the image and after that you're stuck with the colors and symbols you chose.

To get around issues with rasterized tiles, another type of tiles called vector tiles have been increasing in popularity. Vector tiles are similar in concept to rasterized tiles in the sense that they are square packets of information structured in a pyramid motif and sent by the server; however, they contain vector coordinates instead of a picture of the data. This allows the styling to be easily changed. Vector tiles are displayed as client-side graphics, so the client software needs to understand what a vector tile is and how to deal with it. Older mapping software and APIs may not be able to consume vector tiles.

We will talk more about vector tiles in Lesson 5 when we work with Mapbox software, since Mapbox pioneered this format and based their company on it. Esri vector tile support is growing, although it has lagged behind that of Mapbox. Esri vector tile functions are concentrated in their ArcGIS Pro product, which we are not using in this course.

Be aware that all the remaining content in Lesson 4 refers to rasterized tiles, and some of the design and performance considerations discussed may be very different when thinking about vector tiles.

Increasing your instance size for this lesson

Building rasterized cache tiles is CPU and memory-intensive. Your server is making thousands of repetitive map draws, sometimes with a very complex MXD in the background. You can build a cache a lot faster if you assign the tile creation to a powerful machine.

This short-term need for high computing power is a perfect use case for cloud computing. A lot of offices don't have a powerful machine to spare for building tiles (usually their beefiest machine is the server that's already hosting their live apps and web services). In this situation, a server administrator could launch a high memory and/or high CPU instance for just a few hours for the purpose of building tiles. The extra cost is often worth the time savings that it takes to build the cache. Once the tiles are created, the machine can be shut down or scaled down.

For this lesson only, you'll change your ArcGIS Server site to run on a High-Memory Extra Large instance. This costs significantly more than the Medium instance type that you've been using (58 cents per hour) , but it will allow you to work with a complex map document and build cache tiles much faster.

  1. If it is not already stopped, stop your ArcGIS Server site using Cloud Builder.
  2. Log in to the AWS Management Console and display the EC2 page for the region that is running your site.
  3. In the left menu, click Instances.
  4. Right-click your instance and select Instance Settings > Change Instance Type.
  5. Choose r3.xlarge (do not check EBS optimized) and click Apply.

    This Memory Optimized Extra Large instance type has over 30.5 GB of memory and 4 virtual CPU units. For reference, the Medium instance type that you've been using up to this point has 4 GB of memory and 2 vCPUs.
  6. To see the specs for all instance types, check out the Amazon EC2 Instance Types page and drool away. Then take yourself back to reality by viewing the Amazon EC2 Pricing page.
  7. Use Cloud Builder to start your site.

Now that you are running on an instance that costs (at the time of this writing) 58 cents/hour as opposed to 6 cents/hour, it's more important than ever that you remember to stop your site when you are done working on your lesson materials for the day. Also, be sure to set your instance type back to t2.medium after building all your tiles in Lesson 4.