Traffic

By aiguha and dgattey

A project for CS032 to draw maps to screen and give directions. Instantiated by map data in the form of a ways file, nodes file, and index file, as well as an optional traffic server. The server can run a large number of clients with one copy of data in memory.

###Backend Project Components

Autocorrect: dgattey
Stars: aiguha
Bacon: aiguha

###Known Bugs None at this time!

###How to Run

Creating the executables: ant create_exec && ant jar
Running server: bin/trafficServer info/ways.tsv info/nodes.tsv info/index.tsv localhost 9999 10000
Running client: bin/trafficClient localhost 10000 (with --debug flag if you want errors to print)

##Design Details We took our frontend package from Maps and made it the client package. The backend package became the server package. Because of our earlier separation, we were able to simply change the HubController and add a core server package and everything else basically stayed the same. The main package has a Main class that given the appropriate number of flags and the correct arguments, starts up either a ClientApp or a ServerApp. Given our interfaces, everything is pretty seamless.

We have a server, client, data, and main package on the top level. Data is shared by both server and client, so there are connections there. But nothing in the client or server packages import the other's code.

###Client The client simply spawns a GUI and attempts a connection. It starts a timer that attempts a connection every few seconds. If it can't connect, the bar at bottom will be red, all traffic data disappears, and the route disappears. If there's a connection, the bottom bar is green and everything works perfectly. Routing works by clicking two points or entering four streets and pressing the Route button. If nothing is found, the appropriate status will be shown. Otherwise, it'll move you to the start location of the route (and highlight it). Traffic is shown on the map from a range of yellow to red, red being worst traffic. White is no traffic. As you zoom out, small roads (under 3 pixels) will disappear for quicker drawing. Anything outside the bounds of the current view won't even be drawn. Chunks load themselves when the view bounds get close enough to another chunk. Everything goes through the HubController which makes a CommController to communicate with the server. We used Callables to return a value from that thread, and an ExecutorService to run it.

###Server The server consists of five main packages. Four former backend packages: io, autocorrect, kdtree and graph, which manage File IO, autocompletion, nearest neighbor searches and shortest path calculations. These remained almost completely unchanged from Maps, as they were already prepped for multithreading. The graph package involved minor additions to enable realtime traffic-based route calculation. The fifth package is core, which sets up the multithreaded server and handles network communication with all clients and the traffic server.

The server starts handling each client on a new thread, allowing the main thread to stay open for command-line interaction from the administrator (if any). As per our protocol, every request from the client is made on a new socket (except traffic requests) and responded to on the same one. Thus, one client typically has about four sockets connecting to the server. This strategy eliminates the bottleneck of writing multiple responses to one socket concurrently.

When a new socket connects, the request header is read and it is dispatched to one of five response methods in ResponseController, which uses ACController, KDTreeController, GraphController and TrafficController to construct responses.

The server, through TrafficController, is set up to continually try connecting to the traffic server, and constantly read traffic data while connected. When traffic data is unavailable, the server can still perform normally, including route calculation.

###Connections Our client, server, and traffic bot all can be connected together in any order. The client has a timer that attempts connection every few seconds and updates the UI accordingly. Therefore, it doesn't matter if the client starts before or after a server. Similarly, if the server can't send to a request, it removes it from its pool and forgets about it. Additionally, the server attempts reconnection to the traffic server if there's no connection every few seconds, so it can die and be restarted and the server will keep accepting data. The command line interface notifies you about connection status when you run 'status'. If there's no connection, it saves last known traffic data for use when routing. Since the client has no use for the data after a lost connection, it clears its traffic data but keeps its map data.

##Optimizations ###Multithreading

Initial loading of all nodes (for KDTree)
Predictive paging for GUI
Route finding for GUI
20 chunks loading at a time, no more

###Other

Only having one request for a route at a time
Canceling of threads on server and client
Reusing objects as much as possible (HashMaps and HashSets)
Each request runs on its own thread

###Communication Protocol ####Objects

#####Simple Objects

LatLongPoint
Error Messages

These are denoted as #[object_tag]: [fields...] on a single line.

#llp: 41.12 -71.21

#####Complex Objects

List[String]
MapWay
List[MapWay]

These are denoted as <[object_tag]: {optional length}:\n [fields...] >

<way:
/w/0001
Thayer Street
/n/0002
#llp: 41.12 -71.21
/n/0003
#llp: 47.50 -78.401
>

Other objects are similarly constructed.

####Requests And Responses

All queries have the following structure:

@q:query_name:
[object...]
@x

All responses have the following structure:

@r:response_name:
[object...]
@x

##Testing Test our code by running the following:

Server system tests: ant server_test
Client system tests: ant client_test
JUnit tests: ant unit_test
More Server tests start the server on port 9999. Once it's open, run ./testRequests.sh and it should print out responses and expected responses

###By Hand We extensively tested the interactions between the client and server by running netcat and tossing bad data at both sides. Additionally, we started a server and connected five clients to it and did everything we could to make it mad. Lots of long routes, concurrent requests, same location requests, etc. Additionally, we tested closing the connection from the server to the client and vice versa in the middle of a request. No matter what order we did things, nothing failed! Additionally, we modified the traffic bot to send us information every x amount of time, and tested it with requests every 0.0000001 seconds. The whole map filled up, but even with crazy amounts of data coming in, routes were found and reflected the amount of traffic at the moment at which the route was created.

dgattey/traffic

Traffic