Accessing databases through multiple threads

Most applications require store information. The options and scenarios are diverse. Let's focus on the scenario in which multiple remote clients need to store centralized information. No matter if the remote client is a standalone application, website or mobile app

   In general, it is not advisable to let the remote clients communicate with the database directly. It is always better to create a middleware to manage the persistence of the entire system. Not only for safety reasons but because we have a single point of entry to the database so that we can enhance and manage performance in a unified way.

So customers send and receive data from the server and he will be responsible for managing the persistence of this information. In small databases  with few connections, not much to worry about. With notions of SQL you can make a small server with good performance.

But when you are facing scenarios involving large numbers of users or large transactions, you must use more advanced techniques. There are two major problems to deal:

•      Large number of users:

      This is more a question of architecture, but it is always good practice create a new thread for each new connection to a remote client. Of course, you have to limit the number of threads depending on your environment if you want to ensure quality to  everyone connected.
You cannot share the same connection with all the threads or rather, you can do but as database providers implement sessions  in a synchronized way, the commands will not be executed until their turn. You will be within a FIFO (first in, first out) queue.

The idea is to create a new database session for each external connection ( owned by the new thread). Each remote client will get or keep your data in parallel as if he just connected to the database.

So we're done ... isn’t it?

Unfortunately, this is not possible or at least not always. You can’t open a new  session for each incoming connection (and therefore the same number of threads) because sessions database are costly both in terms of resources and licenses (depending on how each provider licensing).

 The solution to this problem is to create a pool of database connections. Each thread will request a free connection pool when needed and released when he finishes what he has to do. Thus, each thread will not own a connection. Actually there is no sense to maintain a connection to the database without using. In short, each thread will request a connection when needed, will do the job, and then release the connection.

…but  it can still happen that when needed, there are no available  sessions  to the database in the pool. In this case, the thread waits until another thread release a database session. In this scenario, the user will likely experience some delay. If this happens frequently, you should ask if you can increase the number of sessions in the pool, or if we have reached the maximum capacity, purchasing more resources to enable your server to handle this load.

In relation to pool connections, there are some providers (providers including databases) that provide implementations that can be used via API. We have used some of them and they work quite well. You also have the option of making your own implementation. It will cost you some time, but on the other hand, you will have more control over what you do and you might particularize and optimize it as it suits us.

If you choose to do it yourself, a good idea is to open the database sessions when the server starts because  the creation of a database connection has a high cost. Therefore, if the connection is opened at start up, there will be no perception of the cost of establishing the connection for remote clients

•   Long transactions or lot of queries / commands involved.

      There are other scenarios or mixed scenarios in which the issue is that you have to run a lot of queries for the same user request. The perception of the users is that your request takes longer than expected. How to deal with this?

Depends deeply on the scenario in which we find ourselves. In fact, to solve this problem requires insight into the performance of your application. We can use a similar strategy to that we used on the last point.

We will create a specific thread for each group of linked commands. Take an example, if you are reading data from a client and their orders, you can think of parallel load data from customer and every order, because, no apparent dependence. So, again, we will create different threads, requesting a connection to the pool, and run these commands in parallel. As a result of this approach, users will wait  only the slowest group of commands that are executed in parallel.

If a branch is heavier than the rest, we can choose to return part of the data while the heavier parts are still running. If you do this, the remote client will need to know in order to inform the user about it. At the end of the heavier parts, send the information to the client so that you can complete the user request.

This approach has some risks that you need to know:

1.   You will  deal with some threads in parallel, so you are in charge of joining together when they finish their duties. Be careful because you have to ensure that the information you are accessing is already loaded. Use semaphores or other mutual exclusion mechanism to control access to these areas.  
2. Beware of deadlocks that can cause connection pool usage. Release the session as long as you do not need it. For example, if you create a new thread from another thread already has a session, the latter may end up locked (depending on the size of the pool and context). You could be waiting for her son before releasing your session while he is also waiting for an  available session. If this behavior is widespread we could have a global lock the entire application. This scenario may seem unlikely, but possible oop often suffer this problem. To address this, as mentioned, let's release the database session when we don't need anymore. But be careful, when dealing with hierarchies of objects and multithreading is not as straightforward ensure this behavior. A simple solution is to restrict parallel to the objects that have no dependency downwards. In a graph of tree , we would only  put in parallel the leaves because they have no dependencies.

Combining this two techniques you can get a good performance on your intensive database access applications.

Building Mobile Games Using the MVC Pattern

When we first decide to create any application we have to keep in mind that our app can be ported or migrated to other platforms. Maybe we will never have to do it but keeping this in mind we will create a better app. So we will need a good layer design to get this goal. A good way is to follow the MVC (model view controller) paradigm. In a few words:

• Model: The logic that remains under any application. Also known as Business Rules.

• View: The tier that paints to the screen.

• Controller: The middle layer that connects Model and View and handles user events by calling the model and displaying to the screen to the View layer.

Keeping these layers separated will allow you to switch your UI framework with less changes than expected.

If we think in terms of oop, we can  implement this architecture  by creating interfaces between view and controller layers. This will force you to keep this model and avoid the temptation to mix these layers. If you are able to abstract the view layer, it will be easier to implement an application based on the MVC architecture

For instance, look at our view interface that we are using to develop our remote client in our   multiplayer game: “Snake the net”…

public interface ClientNibbles {

 public void getSet();

 public void go ();

 public void update (Board game);

public Movement getMovementDirection();

 public int getPlayerId();

 public int getIdGame();

 public int getnumPlayerGame();

public void setPlayerId(int id);

 public void setIdGame(int idGame);

 public void setnumPlayerGame(int NumPlayerGame);

}

There is actually two main operation to implement: update method and getMovementDirection .  We send, each tick, to the view layer a board object with the position of each players. The view layer just needs to read this object and paint a representation of it to the screen.  The Controller and model layers just deal with the server which movements to do but, once commited, they just send a “photo” to the view layer in order to paint the screen. In the reverse way, the controller layer will get own movements by calling getMovementDirection (). The rest of the interface defines control method to configure an incoming game. I think that is an exciting and simple approach. Isn’t it?

If we follow this proposal of architecture we will just need to port the upper (view) layer in some scenarios. In detail:

Our game is intended to be, at first, an Android game. Therefore, we only have the option of using Java as a programming language. You could be tempted to go to Android Studio (or ADT) and start right away. Wait a minute… Test and debug a mobile game is always more complex compared to an standalone app. Thus, basing on our MVC model, we will make fisrt  a standalone  application, fix bugs and once we have stabilized we will just write the part of android view. Think what you will get:

·        

•        Keeping just one source of the code (Controller and logic).

•       Once developed the project in Android, if we detect incidents related to the core It’s easier to come back to standdalone platform to find out the problem instead of using android. Then once  fixed the error, we would  check that everything is ok at mobile scope.

 In our app, we started by implementing a standalone client that implemented the interface defined at view level. This allow us to test the model layer without using any mobile device or simulator, just simple java client apps. By doing this, we are also creating an pure Java game that can be played by his own.

Once stabilized we just need to implement the defined interface (view layer) at android scope. We created dedicated activities, layouts and event custom views to handle this.  Once done, we are able to test the game using two different types of clients: swing java and android clients.

Tests with different remote clients(Java(Android)

But above all, we have to recall that we are designing a multiplayer game. It will be a game that needs someone to accept or discard the movements done by each movement. The asynchronous approach (peer to peer) won’t work since we need someone to decide if a remote movement is on time or late and if we let the clients decide this acceptance or not, it can produce different games in each client… and this would be a disaster. So, we need a mandatory server to handle this. And a server has clients…

So, actually, we have two components to implement: client and server. We also decided to separate these components through interface contracts. So once implemented the server side and deployed where we decided we wouldn't need to change anything (lifecycle apart). While each part implements his side everything will be OK.

Let's turn back at the client side. There are some theories that tell us that remote players in multiplayers games should be  something like a thin client. This means that each client will wait until the server sends the players movement message to paint into the screen. You can imagine how laggy will be perceived by the user. To improve this scenario, and overcome   the perceived latency we decided that clients are going to predict his own movements. We will implement a mandatory server and a client who makes predictions. (The server consolidates and send back to each client the  consolidated movements for each tick.). By doing this, every client can’t move on his own (or almost).

Andoid client screenshot

So the client won't be a thin client. He may also needs to play. Actually, he needs to have the same logic that the server so we will share  the logic code, or almost, by the client and the server. Again we will share the model code between client and server.

There are just Two differences with the server:

• The client does not consolidate any movement. He just makes his own prediction and moves his own players. He will move the rest of  players once received the consolidated movements by the server

• If we are wrong in our prediction we will get back to the last tick  and we will replay our own movement based on the server information.

To deal with these differences you  just need to  create an additional layer in order to handle these differences without affecting the common part.

First Footage

It’s time to make our dreams come true. We have been writing some tech entries about how to do a multiplayer game. Now, you can see how  we have implemented this ideas . Look at the following video: 4 players/snakes online trying to survive:

You can detect easily some client time tick corrections (just notice that client tick is not always the same). In the other hand, it’s not easy to detect that there is one correction at client side. During the game, one movement arrives late to the server so when the consolidated server movements gets back to the client, he’s forced to correct the initial prediction and moves based on server movements. As a user, you will see how an initial movement is quickly modified(our head changes his position).

The best of all is that everything works in a smooth way.

Cheating Latency

In our last posts we have been dedicating our efforts to reduce the client perceived latency. Let’s check what   we have  focused:

-          We have configured TCP to flush data as soon as it has been generated.

-          We decided to mix TCP & UDP whenever is possible.

-          We create control layers to sync client and server to get a smooth gameplay.

-          We made our own Object Serialization .

All points above are about  technical issues. Now, let’s turn inside our app and let's check if we can enhance its performance.

Summarizing how the server works  will show us some clues about which point can be improved. Let’s look inside:

-        The server and the remote clients wait, initially,  the same gap of time (tick) for each turn.

-          The server wakes up each time tick and consolidates the players movements sent previously by the remote clients.

-        The server sends back a message containing the consolidated movements and tells every client if they have advanced or delayed related to the objective lag between them.

-          The remote clients make some adjustments on the client side to compensate changing latency related to the server.

-          In order to make this work, clients begin their tick before the server in order to compensate each own latency plus certain buffer.

Now, let’s think what we can do to improve this algorithm:

First Enhancement: Try to send  consolidated movements once all player movements have arrived.

As we have seen, the  server  waits until the end of the tick to send back consolidated movements to the client. That’s how it should be, you may think. Sure? Is it strictly mandatory to wait until the end of the server tick if we already have all movements?

Let’ see... We have nothing to do once all movements are received, at server side, except waiting to the conclusion of the tick. So, if we advance the message containing the consolidated movements,  remote  clients will receive sooner that movements and they will paint non own movements before.

As a result, remote clients will see a more fluid and dynamically game and that’s one of our main goals.

            Risks? If the end client  tick time is so close  to the server end tick , we will be in risk of arriving late for the following movements. That’s where our algorithm, described in our last post, works. The following client tick will be shorter than usual in order to keep the same gap related to the server. If it happens very often  we will only feel a faster game and, in fact, this is what we want. Anyway, we always can wait a minimum of time before sending back consolidated movements. In our case, Snake the Net, we have not implemented this guarantee, except at LAN modes,  because we want a fast game and unfortunately, the latency itself introduces more time that we wanted at minimum.

We implemented this approach with sensible results and I recommend to apply similar strategies wherever  is possible.

Second Enhancements: Extra time for the lazy clients

  Let’s think in the opposite way. What if we have lazy clients? What if we add some extra milliseconds to the server tick and then we tell him to recover that time by reducing his tick client? By doing this, we avoid to correct our clients despite he is a really lazy client. We make him to do a fast tick but, in the other hand, we avoid to correct him. In our case this enhancement  help  us by reducing corrections and improving stability. Of course, you have to control if a lazy client is actually a disconnected client. Again, sense common. If he is late several consecutive times, we will disconnect him and replaced by a bot, that they always arrive at time ;-)