MySQL JPQL query performance issue

Question

I'm building a game which I intend to be a persistent online world (like an MMO, but probably not a massive number of players.) I'm making a game map by just creating a 2d grid of "tiles." I've got my tiles persisting to my database just fine, and I've got a web page UI (via Vaadin) set up so I can view them just fine. So far I've been viewing them most often in a 20x20 grid.

Up to now, it seemed to be going great. I could persist a couple sets of 10,000 tiles, and scroll around looking at them and redisplaying the tiles was nearly instantaneous. Today, though, I decided to persist a whole bunch of tiles (633,460 to be exact) to see what it was like with a more realistic number of tiles in the database. Now I'm having some significant performance issues scrolling around the map, because my MySQL database seems to be having a difficult time handling that many tiles.

I'm using JPA 2.0 backed by Hibernate. Even when I go straight to the database and run a SQL query, the response times are super slow.

SELECT * FROM game.landtile WHERE xcoor < 999999061 AND xcoor >  999999040
AND ycoor > 1000000140 AND ycoor < 1000000161;

I just ran that query to get 400 landtiles and it had a 3.5 second duration and 7.25 second fetch time directly in MySQL! If I run the same query a second time it has a 0.5 second duration and a 0.1 second fetch time. 500 miliseconds I could live with ... 7,000 I cannot.

My entity for the landtile looks like this:

@Entity
@Table(name = "LandTile", uniqueConstraints = {@UniqueConstraint(columnNames={"XCoor", "YCoor"})})
public class LandTile extends AbstractMapValues implements Serializable {

    private static final long serialVersionUID = 1L;

    @Id
    @Column(name = "LandTileId")
    private String landTileId;

    @Column(name = "XCoor")
    @Index(name = "XCoordinateIndex")
    @Range(min = 0, max = MAX_MAP_INT)
    private int xCoor;


    @Column(name = "YCoor")
    @Index(name = "YCoordinateIndex")
    @Range(min = 0, max = MAX_MAP_INT)
    private int yCoor;

    @OneToOne(cascade = CascadeType.ALL)
    private Building building;

    @OneToOne(cascade = CascadeType.ALL)
    private Road road;

    private boolean river;

    @OneToMany(mappedBy="expeditionLocationLandTile")
    private Set<Expedition> expeditionsAtThisLocation;

    @ManyToMany(
            mappedBy = "discoveredLandTiles",
            targetEntity = Player.class
        )

    private Set<Player> playersThatHaveDiscovered;

    @ManyToMany(
            mappedBy = "routeListOfLandTiles",
            targetEntity = Expedition.class
        )
    private List<Expedition> expeditionRouteThisIsPartOf;

    @ManyToOne
    @JoinColumn(name="FiefdomOfCapitalCity", unique=true)
    private Fiefdom fiefdomOfCapitalCity;

    @ManyToOne
    @JoinColumn(name="ActualOwningFiefdomId", unique=true)
    private Fiefdom actualOwningFiefdom;

    @ManyToOne
    @JoinColumn(name="DefaultOwningFiefdomId", unique=true)
    private Fiefdom defaultOwningFiefdom;

There are also another 18 simple int columns I didn't list here.

So basically I'm hoping you guys could help me figure out what my options are. As you can see there are a lot of relationships going on between a landtile and my other models, and for a game like this the whole data set itself is highly relational, which is why I went with a SQL database in the first place. But now I'm worried the performance is just going to be so bad that I should stop developing new features, and rework my entire DAO layer (hopefully it would only have to be that ...) to use a NoSQL database, and deal with all my relationships in the java.

Looks like MongoDB has built in support for 2d grids:

http://blog.codecentric.de/en/2012/02/spring-data-mongodb-geospatial-queries/

I was planning on probably hosting my game on Amazon and using Amazon Elasti-cache. Would it be possible for my game map data to mostly linger in that 2nd level cache (via hibernate) and therefore I wouldn't have this horrible performance? I really think the only real performance issue I'll run into with MySQL is going to be scrolling around the dumb map!

Answer 1

My data is spatial data, and MySQL's InnoDB engine does not support spatial indexes. The MyISAM engine does, but it does not support transactions or foreign keys. See the MySQL documentation .

Instead of sacrificing transactions and foreign keys, I decided to use a database that supports complex relationships, transactions, and foreign keys, so I switched my entire project to use the Neo4j graph database .