Tuesday, November 25, 2008

KISS Architecture Mobile Sample This is the final post in this series, and for a complete reference, here are the previous parts:

  • Part 1 was a general introduction
  • Part 2 outlined the architecture (tiers, etc)
  • Part 3 showed the benefit of loosely coupled tiers (distribution, cloud, etc)
  • Part 4 started the implementation by creating the entity data model (using ADO.NET Entity Framework)
  • Part 5 published the entity data model as a data service (using ADO.NET Data Services)
  • Part 6 implemented the business domain (using the data service)
  • Part 7 created the service (using WCF)
  • Part 8 started the implementation of the mobile client (using WCF)
  • Part 9 added offline support to the mobile client (using SQL Server Compact, etc)

The implemented architecture is published on CodePlex in a project called KISS Architecture, and this means that you can access the full source code as well as discuss it, come with suggested improvements, etc.

On the upper right you see the UX of the sample client included in the architecture sample, and the functionality is that a part of a city name can be entered in the text box, and when the "Get" menu item is selected, the grid is filled with the matching customers.

Even if this series is complete, I will continue to build further on this architecture blueprint, and any suggestions on things to add are most welcome. Any other feedback, for that matter, is also welcome!

posted on Tuesday, November 25, 2008 10:05:50 PM UTC  by Chris  #    Comments [0]
 Friday, November 21, 2008

In this part, I will continue the implementation of the presentation tier for Windows Mobile, and if you want some background, please see the previous parts:

  • Part 1 was a general introduction
  • Part 2 outlined the architecture (tiers, etc)
  • Part 3 showed the benefit of loosely coupled tiers (distribution, cloud, etc)
  • Part 4 started the implementation by creating the entity data model (using ADO.NET Entity Framework)
  • Part 5 published the entity data model as a data service (using ADO.NET Data Services)
  • Part 6 implemented the business domain (using the data service)
  • Part 7 created the service (using WCF)
  • Part 8 started the implementation of the mobile client (using WCF)

The implemented architecture is published on CodePlex in a project called KISS Architecture, and this means that you can access the full source code as well as discuss it, come with suggested improvements, etc. As I walk you through the creation of the architecture, I suggest you keep the source code handy to check out more details.

In the previous part (8) of this blog series, I showed how the Windows Mobile client can call a WCF service when connected to the network. However, there are a number of times when the mobile device is not connected and still need to allow the application to function. It would be really awesome if the new technologies like ADO.NET Data Services, ADO.NET Entity Framework, LINQ to Entities, etc, were available also for Windows Mobile and .NET Compact Framework, as this would mean that we could use identical code (assemblies/tiers) on the client, but unfortunately they are not (yet).

So, what is the second best solution? I still want to make use of the entities defined by the (data) service tier, and not maintain any custom entity code. Also, I want the implementation to be as simple as possible (remember the KISS principle) and make it as easy as possible to upgrade when these new technologies are available. That means that I don't want to spend time writing my own custom framework, but rather something simple that just works.

To allow local data storage, I create a SQL Server Compact database by adding a "Database File" named Northwind.sdf to the Kiss.Mobile project. To that database, I then add the same tables that our order (data) service defines (see part 4 of this blog series). I also add a plain class named Common.cs to the same project that is a singleton holding a single connection to the database.

With the local database in place and a connection to it, the code for the "Get" menu item can be changed into:

private void getMenuItem_Click(object sender, EventArgs e)
{
   
Customer[] customers;
   
if(online)
    {
       
OrderServiceClient.EndpointAddress = new EndpointAddress("http://192.168.0.100:2222/OrderService.svc");
       
OrderServiceClient service = new OrderServiceClient();
        customers = service.GetCustomersByCity(cityTextBox.Text);
    }
   
else
    {
       
OrderHandler data = new OrderHandler();
        customers = data.GetCustomersByCity(cityTextBox.Text);
    }
    dataGrid.DataSource = customers;
}

The variable indicating connection state (online) can be either manual or automatic through SystemState.ConnectionsCount. Before we take a look at the OrderHandler class, here's the updated code for the "Update" menu item:

private void updateMenuItem_Click(object sender, EventArgs e)
{
   
Customer customer = ((Customer[])dataGrid.DataSource)[0];
    customer.City = customer.City + "X";

   
if(online)
    {
       
OrderServiceClient service = new OrderServiceClient();
        service.UpdateCustomer(customer);
    }
   
else
    {
       
OrderHandler data = new OrderHandler();
        data.UpdateCustomer(customer);
    }
}

Ok, so with that in place, here's the implementation of the OrderHandler class:

public class OrderHandler
{
   
public Customer[] GetCustomersByCity(string city)
    {
       
//var q = from c in data.CustomerSet
        //        where c.City.Contains(city)
        //        select c;
        return HandlerHelper.GetObjects<Customer>(string.Format(
           
"SELECT * FROM Customers WHERE City LIKE '%{0}%'", city));
    }

   
public void UpdateCustomer(Customer customer)
    {
       
HandlerHelper.UpdateObject<Customer>(customer, string.Format(
           
"SELECT * FROM Customers WHERE CustomerID='{0}'", customer.CustomerID));
    }
}

Not much happening here except that the handler make use of a helper that I wrote to take care of the plumbing. I can use the entity type and I need to define the query as plain SQL, but note that I've kept the desired LINQ as a comment in preparation for the day when it can be used. Let's look at the implementation of the helper class:

public static class HandlerHelper
{
   
public static T[] GetObjects<T>(string sql) where T : new()
    {
       
List<T> objects = new List<T>();
       
SqlCeCommand command = Common.Values.DatabaseConnection.CreateCommand();
        command.CommandText = sql;
       
SqlCeDataReader reader = command.ExecuteReader();
       
Type type = typeof(T);
       
while(reader.Read())
        {
            T entity =
new T();
           
for(int i = 0; i < reader.FieldCount; i++)
            {
                PropertyInfo propertyInfo = type.GetProperty(reader.GetName(i));
                propertyInfo.SetValue(entity, reader[i],
null);
            }
            objects.Add(entity);
        }
       
return objects.ToArray();
    }

   
public static void UpdateObject<T>(T entity, string sql)
    {
       
SqlCeCommand command = Common.Values.DatabaseConnection.CreateCommand();
        command.CommandText = sql;
       
SqlCeResultSet resultSet = command.ExecuteResultSet(ResultSetOptions.Updatable);
        resultSet.Read();
       
Type type = typeof(T);
       
for(int i = 0; i < resultSet.FieldCount; i++)
        {
           
PropertyInfo propertyInfo = type.GetProperty(resultSet.GetName(i));
           
object value = propertyInfo.GetValue(entity, null);
            if(!value.Equals(resultSet.GetValue(i)))
                resultSet.SetValue(i, value);
        }
        resultSet.Update();
    }
}

The database connection kept by the singleton (Common.Values) is used to either query or update the local database. Both methods are implemented using generics, and reflection is used to find all the attributes (fields). To optimize performance, a data reader is used for the query and a resultset is used for the update. Both methods are somewhat simplified as some extra null handling is necessary but does not add to the discussion. Using reflection is not an ideal solution, but remember that my goal is to keep the implementation as simple as possible and also that this is not a solution for the future, it's a temporary fix in the lack of the right technologies on the device. When that day comes, the helper class can be scrapped, and the handler class either updated or even replaced by something like LINQ to Entities.

In the next part, I will conclude this series.

posted on Friday, November 21, 2008 10:04:02 PM UTC  by Chris  #    Comments [0]
 Tuesday, November 18, 2008

In this part, I will start the implementation of the presentation tier for Windows Mobile, and if you want some background, please see the previous parts:

  • Part 1 was a general introduction
  • Part 2 outlined the architecture (tiers, etc)
  • Part 3 showed the benefit of loosely coupled tiers (distribution, cloud, etc)
  • Part 4 started the implementation by creating the entity data model (using ADO.NET Entity Framework)
  • Part 5 published the entity data model as a data service (using ADO.NET Data Services)
  • Part 6 implemented the business domain (using the data service)
  • Part 7 created the service (using WCF)

The implemented architecture is published on CodePlex in a project called KISS Architecture, and this means that you can access the full source code as well as discuss it, come with suggested improvements, etc. As I walk you through the creation of the architecture, I suggest you keep the source code handy to check out more details.

kisswmform

I start the implementation of a Windows Mobile client by creating a new project of type "Smart Device Project", name it Kiss.Mobile, select "Windows Mobile 5.0 Smartphone SDK", ".NET Compact Framework Version 3.5", and "Device Application". This will allow it to run on many of the devices on the market, and still use the latest framework with new functionality. That new functionality include WCF, so let's move on by creating a reference to the WCF service I created in the previous part (7). You do that by using the tool NetCFSvcUtil that is included in the Power Toys for .NET Compact Framework 3.5. For more details on using this tool, please see A New Mobile N-tier Architecture (part 8).

Let's implement a simple user interface by renaming the automatically generated form (Form1) to MainForm, and add three controls (a Label named cityLabel, a TextBox named cityTextBox, and a DataGrid named simply dataGrid) to it. Also, add a MainMenu named mainMenu and add menu options to it named doneMenuItem, getMenuItem, and updateMenuItem. The result should look similar to the figure on the left.

With the service reference and the user interface in place, it's time to implement some presentation logic. Here's the code for the "Get" menu item:

private void getMenuItem_Click(object sender, EventArgs e)
{
    OrderServiceClient.EndpointAddress = new EndpointAddress("http://192.168.0.100:2222/OrderService.svc");
   
OrderServiceClient service = new OrderServiceClient();
    Customer[] customers = service.GetCustomersByCity(cityTextBox.Text);
    dataGrid.DataSource = customers;
}

Note how easy it would be to redirect the client to use the service in another location (other server, cloud, etc) by just changing the service URI. Also note how the definition of the entity (Customer) have come all the way from the data service without any manual coding anywhere, and when it is changed, it's simply a matter of updating the service references in all tiers. The observant will note that the List<Customer> return value of the service is transformed into a plain array when serialized by WCF, but if a list is preferred, this code could be used:

List<Customer> customers = service.GetCustomersByCity(cityTextBox.Text).ToList();

The code for the "Update" menu item looks like this:

private void updateMenuItem_Click(object sender, EventArgs e)
{
   
Customer customer = ((Customer[])dataGrid.DataSource)[0];
    customer.City = customer.City + "X";
    OrderServiceClient service = new OrderServiceClient();
    service.UpdateCustomer(customer);
}

It first takes the first row in the grid, and adds an "X" to the City attribute. Then the service is called with the updated entity instance (customer). It can't be much simpler than this!

In the next part, I will continue the implementation of the Windows Mobile client by adding offline support.

posted on Tuesday, November 18, 2008 10:02:41 PM UTC  by Chris  #    Comments [0]
 Friday, November 14, 2008

In this part, I will continue the implementation of the middle tier by publishing a WCF service, and if you want some background, please see the previous parts:

  • Part 1 was a general introduction
  • Part 2 outlined the architecture (tiers, etc)
  • Part 3 showed the benefit of loosely coupled tiers (distribution, cloud, etc)
  • Part 4 started the implementation by creating the entity data model (using ADO.NET Entity Framework)
  • Part 5 published the entity data model as a data service (using ADO.NET Data Services)
  • Part 6 implemented the business domain (using the data service)

The implemented architecture is published on CodePlex in a project called KISS Architecture, and this means that you can access the full source code as well as discuss it, come with suggested improvements, etc. As I walk you through the creation of the architecture, I suggest you keep the source code handy to check out more details.

It's time to create the service, and I do that by creating a new project of type "WCF Service Application" and name it Kiss.Service. First I add a reference to the Kiss.Domain project, and then I rename the automatically generated interface to IOrderService.cs. I change its code to...

namespace Kiss.Service
{
    [
ServiceContract]
   
public interface IOrderService
    {
        [
OperationContract]
       
List<Customer> GetCustomersByCity(string city);

        [
OperationContract]
       
Customer GetCustomer(string id);

        [
OperationContract]
       
void UpdateCustomer(Customer customer);
    }
}

...and make sure all references to it is refactored (there are a number of places, like in the Web.config file). Then I rename the generated service class to OrderService.cs and update its (references and) code to:

namespace Kiss.Service
{
    [
AspNetCompatibilityRequirements(RequirementsMode = AspNetCompatibilityRequirementsMode.Allowed)]
   
public class OrderService : IOrderService
    {
       
private OrderDomain orderDomain = new OrderDomain();

       
public List<Customer> GetCustomersByCity(string city)
        {
           
return orderDomain.GetCustomersByCity(city);
        }

       
public Customer GetCustomer(string id)
        {
           
return orderDomain.GetCustomer(id);
        }

       
public void UpdateCustomer(Customer customer)
        {
            orderDomain.UpdateCustomer(customer);
        }
    }
}

This is the actual implementation of the service that implement the interface mentioned above, and each method simply use the domain class to do its work. The decoration (attribute) of the service class is needed to make the service available for a Silverlight 2 client. Even if no such client exist for Windows Mobile yet, this is a good measure to prepare for the future.

Another similar measure is to include two more files in the service project, and the first is clientaccesspolicy.xml and the second is crossdomain.xml. The purpose of these files is to make the service available for applications that make cross domain calls, and the first is specific to Silverlight and the second is used by Flash applications, among others. Here is a sample clientaccesspolicy.xml file...

<?xml version="1.0" encoding="utf-8" ?>
<
access-policy>
    <
cross-domain-access>
        <
policy>
            <
allow-from http-request-headers="*">
                <
domain uri="*" />
            </
allow-from>
            <
grant-to>
                <
resource path="/" include-subpaths="true" />
            </
grant-to>
        </
policy>
    </
cross-domain-access>
</
access-policy>

...that allow any client to access the service. A final measure that is necessary for both Silverlight and Windows Mobile WCF clients is to make sure that the service use the basicHttpBinding, and here is an extract of the Web.config file:

<system.serviceModel>
    <
services>
        <
service behaviorConfiguration="Kiss.Service.OrderServiceBehavior" name="Kiss.Service.OrderService">
            <
endpoint address="" binding="basicHttpBinding" contract="Kiss.Service.IOrderService">

In the next part, I will start implementing the Windows Mobile client.

posted on Friday, November 14, 2008 10:01:15 PM UTC  by Chris  #    Comments [0]
 Tuesday, November 11, 2008

In this part, I will start the implementation of the middle tier, and if you want some background, please see the previous parts:

  • Part 1 was a general introduction
  • Part 2 outlined the architecture (tiers, etc)
  • Part 3 showed the benefit of loosely coupled tiers (distribution, cloud, etc)
  • Part 4 started the implementation by creating the entity data model (using ADO.NET Entity Framework)
  • Part 5 published the entity data model as a data service (using ADO.NET Data Services)

The implemented architecture is published on CodePlex in a project called KISS Architecture, and this means that you can access the full source code as well as discuss it, come with suggested improvements, etc. As I walk you through the creation of the architecture, I suggest you keep the source code handy to check out more details.

With the data tier in place, I'm ready to create the middle tier, and I start by creating a plain class library named Kiss.Domain which will hold my business domain classes. I add a service reference to the data service created in the previous part (5) of this blog series, and get the complete entity data model through the generated proxy class. Then I create a plain class named OrderDomain.cs and add the following code to it:

namespace Kiss.Domain
{
   
public class OrderDomain
    {
       
OrderEntities data = new OrderEntities(new Uri("http://localhost:1111/OrderDataService.svc"));

       
public List<Customer> GetCustomersByCity(string city)
        {
           
var q = from c in data.CustomerSet
                   
where c.City.Contains(city)
                   
select c;

           
return q.ToList();
        }

       
public Customer GetCustomer(string id)
        {
           
return data.CustomerSet.SingleOrDefault(c => c.CustomerID == id);
        }

       
public void UpdateCustomer(Customer customer)
        {
            data.AttachTo(
"CustomerSet", customer);
            data.UpdateObject(customer);
            data.SaveChanges();
        }
    }
}

It implements three common methods for getting a list of customers for a specified city, getting a specific customer by identity, and for updating a customer. Note that we in the list query method (GetCustomersByCity) can use LINQ to query the data service, and if you debug the code and check out the value of the query variable (q), you will see the actual query (REST) URI generated. LINQ is also used in the second query method (GetCustomer), but here in an alternative format (and with a lambda expression). Note also how simple the code is to update a customer using the data service. I have intentionally left out the other CRUD operations (insert and delete) as they should not be to hard to figure out, and don't add much to the architectural discussion.

With this in place, you see how easy it would be to move the data service to another location (like the cloud), by simply changing the URI. Likewise, the switch between online (to the server or cloud) and offline (to a local data service) would be as easy. Unfortunately, as the ADO.NET Data Services is not yet available on Windows Mobile (yet), we need to accomplish something similar in another way (which I will show in a later part of this blog series).

In the next part, I will implement the services tier.

posted on Tuesday, November 11, 2008 9:59:15 PM UTC  by Chris  #    Comments [0]
 Friday, November 07, 2008

In this part, I will continue with the implementation of the data service using ADO.NET Data Services, and if you want some background, please see the previous parts:

  • Part 1 was a general introduction
  • Part 2 outlined the architecture (tiers, etc)
  • Part 3 showed the benefit of loosely coupled tiers (distribution, cloud, etc)
  • Part 4 started the implementation by creating the entity data model (using ADO.NET Entity Framework)

The implemented architecture is published on CodePlex in a project called KISS Architecture, and this means that you can access the full source code as well as discuss it, come with suggested improvements, etc. As I walk you through the creation of the architecture, I suggest you keep the source code handy to check out more details.

Now it's time to publish the entity data model (OrderModel.edmx) that we created in the last part (4) of this blog series, and that is done by adding an "ADO.NET Data Service" named OrderDataService.svc to the data tier assembly, Kiss.Data. The modifications necessary to this service class is trivial, and involves specifying the type of the data model (object context, in my case OrderEntities) and making the complete model available to anyone. The resulting code looks like this:

namespace Kiss.Data
{
   
public class OrderDataService : DataService<OrderEntities>
    {
       
public static void InitializeService(IDataServiceConfiguration config)
        {
            config.SetEntitySetAccessRule(
"*", EntitySetRights.All);
        }
    }
}

Amazing as it sounds, the data tier is now ready to be used, and we can start querying the database through the newly built data service. For example, if we go to http://localhost:1111/OrderDataService.svc/CustomerSet('ALFKI'), we get the following result (somewhat simplified)...

<?xml version="1.0" encoding="utf-8" standalone="yes" ?>
<
entry xmlns="http://www.w3.org/2005/Atom">
    <
id>http://localhost:1111/OrderDataService.svc/CustomerSet('ALFKI')</id>
    <
title type="text" />
    <
updated>2008-11-25T02:24:45Z</updated>
    <
author>
        <
name />
    </
author>
    <
link rel="edit" title="Customer" href="CustomerSet('ALFKI')" />
    <
link type="application/atom+xml;type=feed" title="Orders" href="CustomerSet('ALFKI')/Orders" />
    <
category term="Kiss.Data.Customer" />
    <
content type="application/xml">
        <
properties>
            <
CustomerID>ALFKI</CustomerID>
            <
CompanyName>Alfreds Futterkiste</CompanyName>
            <
ContactName>Maria Anders</ContactName>
            <
ContactTitle>Sales Representative</ContactTitle>
            <
Address>Obere Str. 57</Address>
            <
City>Berlin</City>
            <
Region null="true" />
            <
PostalCode>12209</PostalCode>
            <
Country>Germany</Country>
            <
Phone>030-0074321</Phone>
            <
Fax>030-0076545</Fax>
        </
properties>
    </
content>
</
entry>

...and this allows us to query the complete data model through a RESTful interface. Note that you may have to turn off the automatic feed formatter included in Internet Explorer 7 to see the resulting XML (in ATOM format), and you do that by selecting Tools > Internet Options > Content > Feeds Settings, and uncheck "Turn on feed reading view".

In the next part, I will start implementing the middle tier.

posted on Friday, November 07, 2008 9:57:10 PM UTC  by Chris  #    Comments [0]
 Friday, October 31, 2008

In this part, I will get more practical by showing how the first parts of this architecture looks in code, and if you want some background, please see the previous parts:

  • Part 1 was a general introduction
  • Part 2 outlined the architecture (tiers, etc)
  • Part 3 showed the benefit of loosely coupled tiers (distribution, cloud, etc)

As I mentioned in the first part, the implemented architecture is published on CodePlex in a project called KISS Architecture, and this means that you can access the full source code as well as discuss it, come with suggested improvements, etc. As I walk you through the creation of the architecture, I suggest you keep the source code handy to check out more details.

Ok, let's start building the architecture from the ground up! To make things really simple (remember the KISS principle), I start with an "ASP.NET Web Application" named Kiss.Data as the data tier. In that assembly I add an "ADO.NET Entity Data Model" class named OrderModel.edmx, and add the following tables from the Northwind database (renamed Customer, Order, OrderDetail, Product and Category):

KISS Architecture Entity Model

If we take a look at a simplified version of the generated code (in OrderModel.Designer.cs)...

namespace Kiss.Data
{
   
public partial class OrderEntities : ObjectContext
    {
       
public OrderEntities(string connectionString) : base(connectionString, "OrderEntities")
        {
           
this.OnContextCreated();
        }
       
public ObjectQuery<Customer> CustomerSet
        {
           
get
            {
               
if(this._CustomerSet == null)
                   
this._CustomerSet = base.CreateQuery<Customer>("[CustomerSet]");
               
return this._CustomerSet;
            }
        }
       
private global::System.Data.Objects.ObjectQuery<Customer> _CustomerSet;
    }

    [
EdmEntityTypeAttribute(NamespaceName = "Kiss.Data", Name = "Customer")]
    [
DataContractAttribute(IsReference = true)]
    [
Serializable()]
   
public partial class Customer : EntityObject
    {
        [
EdmScalarPropertyAttribute(EntityKeyProperty = true, IsNullable = false)]
        [
DataMemberAttribute()]
       
public string CustomerID
        {
           
get { return this._CustomerID; }
           
set
            {
               
this.OnCustomerIDChanging(value);
               
this.ReportPropertyChanging("CustomerID");
               
this._CustomerID = StructuralObject.SetValidValue(value, false);
               
this.ReportPropertyChanged("CustomerID");
               
this.OnCustomerIDChanged();
            }
        }
       
private string _CustomerID;
       
partial void OnCustomerIDChanging(string value);
       
partial void OnCustomerIDChanged();

...we see that in the model, or object context, (OrderEntities) there is a class created for each entity (Customer) and a query (set) for that entity (CustomerSet). Also note that each of the entity classes and its members has the necessary decorations (attributes) to be serialized correctly over WCF.

In the next part, I will look at how to publish the data model as a data service.

posted on Friday, October 31, 2008 8:54:44 PM UTC  by Chris  #    Comments [0]
 Friday, October 24, 2008

I'm continuing the series that started with part 1 and part 2 by talking more about the changes in a modern mobile multi-tier architecture. This time I will talk about the benefits of having loose coupling of the tiers.

For years, actually since Web Services started to be used, we have had a loose coupling between the presentation and middle tier. The effect has been that the system can be physically partitioned on different machines and even on either side of a firewall. Effectively creating many options for separating presentation from services, like using external services or even outsourcing services. It also allowed the outsourcing of the presentation tier unrelated to the location of services.

With the data tier, it was more difficult. Even if many applications also physically separated the services from the data tier, it was often problematic due to security issues (like opening up odd ports in the firewall to allow database traffic). With a technology like ADO.NET Data Services, it becomes easier to deploy the data tier separated from the services tier. That is not only a benefit for deployments on the server side, but it also opens up the ability to use an external data provider.

One interesting initiative by Microsoft is what they call services in the "cloud" (i.e. on the Internet), known as Azure (note the color of the cloud in the diagram figure below), and it means that we will have a new option for outsourcing the different tiers. With an upside-down view of the tiers (users at the bottom and the data tier on top), the following figure shows a number of available options:

KISS Architecture Tier Distribution

Starting from the left, we have a stand-alone application (A) with all three tiers running locally on the client. The next, is the traditional client/server application (B) with a fat client only hosting the data on the server. This option has not been used much the last couple of years, mostly because of the lack of simple data access, but will probably be more common in application using Silverlight 2 and ADO.NET Data Services. Next, is the typical fat client application (C) using server-side services, and then we have a normal Web application (D). The first option that use the cloud (E) is an application that make use of a cloud-hosted data service (like SQL Data Services), and the last two options also host the services (F) and even the presentation (G) tier in the cloud (using .NET Services). Of course, several of these options can be combined, and an example is an application hosted in the cloud (G) that can also be run offline (A) when the client doesn't have a network connection (a common scenario for a mobile application).

In the next part, I will start implementing the architecture that enable these options.

posted on Friday, October 24, 2008 8:52:17 PM UTC  by Chris  #    Comments [0]
 Friday, October 17, 2008

To continue this series that started with KISS Architecture (part 1), I will now talk more about a modern mobile multi-tier architecture. Because a picture always says more than a thousand words, let's start with a classic one.

KISS n-tier architecture

This image should be familiar to anyone who read the Application Architecture for .NET from back in 2002. Interestingly, the Microsoft's patterns & practices team are working on a new version of this guide called Application Architecture Guide 2.0 that is already in its second beta (there is also a specific pocket guide for mobile architectures). Building on the discussions in that guide, I have taken the old diagram (because I like the colors) and created an updated version with a small modification. The modification is related to the placing of the business entities, that I think belong in the data tier. The main reason for that is that the ORM technologies (in my case, LINQ to Entities or ADO.NET Entity Framework) take care of defining the business entities. For more details on multi-tier architectures, please read the guide, it's a great resource.

A nice addition to the traditional diagram is to show how other systems ("External Systems") access my system's services that are published as service interfaces with WCF, and that is exactly how other systems ("Services") are accessed by my system.

However, the biggest changes compared to my previous architecture blueprint are probably in the data tier. Even if I was a big fan of LINQ to SQL, the ADO.NET team have announced that the long-term plan is that in ".NET 4.0, LINQ to Entities will be the recommended data access solution for LINQ to relational scenarios". Therefore, I now use LINQ to Entities (ADO.NET Entity Framework) in the data tier.

However, the changes in the data tier doesn't stop there. Another (probably even bigger) change is that the data tier is now made loosely coupled for the first time. Many attempts have been made in this direction (SQLXML to mention one) before, but I think that this time Microsoft got it right with ADO.NET Data Services. In short, it's a way of accessing data through a simple RESTful interface over HTTP using formats like ATOM and JSON.

Even if this means that it becomes very easy to access the data source directly form the presentation tier, it doesn't mean that a sound architecture should use or even suggest such nonsense. It may be interesting and fun for quick demos, but doesn't belong in any serious enterprise solution - mobile or not. Therefore, the thinking that I introduced in A New Mobile N-tier Architecture (part 2) about keeping the data and logic as logically bound together as possible (in what I call domains) still holds. I strongly believe that the data should only be accessed through the logic.

In the next part I will talk about the benefits of having tiers that are loosely coupled.

posted on Friday, October 17, 2008 8:49:37 PM UTC  by Chris  #    Comments [0]
 Friday, October 10, 2008

In March, when I started to create a new architecture for building mobile applications for Windows Mobile called Windows Mobile Architecture Blueprint, I thought that it would be possible to use the result for a while. I was both right and wrong. I was right about some of the theory and some of the technologies, but this fall (specifically this year's PDC) has changed a lot of things. I therefore realize that it's already time for yet another attempt to create an architecture blueprint. As last time, I will start with the server side (remember, almost no mobile applications live alone), and then work my way all the way out to the Windows Mobile client.

As you may recall from the last attempt, I'm a big fan of the KISS principle. So big, in fact, that this time I have named the architecture after this principle. The reason is that I think it's of absolute necessity that the architecture is as simple as possible (but no simpler, as Einstein said). Otherwise it will not be used, and the whole point of thinking through and implementing an architecture is that someone will actually use it. Some of the other principles that also apply are YAGNI (don't implement it until you really need it) and DRY (learn to master the difficult art of partitioning a system).

Here are some of the technologies and products that will be used in (or suitable for) the architecture:

  • .NET Framework - the treasure chest
  • C# - my language of choice
  • Visual Studio - the worlds greatest development tool
  • SQL Server - my database of choice (on both server and device)
  • ADO.NET Entity Framework and LINQ to Entities - the way forward on ORMs (for the ADO.NET team)
  • ADO.NET Data Services - how to make data access loosely coupled
  • WCF - on discussion, this is the RPC of our time
  • WinForms/ASP.NET - traditional ways of designing UXs
  • WPF/Silverlight - the new way of designing UXs
  • Azure/Live services - a new alternative to host data, services, and applications

To get some theory and background to the use of some of the above mentioned technologies, I recommend you read the beginning of the last series (you find links to all the blog posts in the last post). This time, I will focus only on the technologies that changed since the last time, and therefore, as an example I will not talk much about WCF and how to access WCF services from a .NET CF application (covered in parts 3, 7, and 8 in the last series).

Just as last time, this effort will be an open source venture hosted on CodePlex with the name KISS Architecture. This means that you can access the full source code as well as discuss it, come with suggested improvements, etc. As I walk you through the creation of the architecture, I suggest you keep the source code handy to check out more details.

Accompanying the source code will be a series of blog posts starting with this one.

posted on Friday, October 10, 2008 8:47:05 PM UTC  by Chris  #    Comments [0]
 Wednesday, May 28, 2008

blueprintuiThis is the final post in this series, and for a complete reference, here are the previous parts:

  • Part 1 was a general introduction
  • Part 2 talked about the changes in the lower tiers (logic + data, LINQ2SQL)
  • Part 3 discussed the changes in communication (WCF)
  • Part 4 covered important stuff in the user interface (MVC)
  • Part 5 summarized the theory and outlined the new architecture
  • Part 6 started the walkthrough of the architecture code by looking at the business domain
  • Part 7 continued the code walkthrough with a look at the service (WCF)
  • Part 8 covered the consumption of the service with .NET CF
  • Part 9 showed the implementation of the user interface (MVC)

The implemented architecture is published on CodePlex in a project called Windows Mobile Architecture Blueprint, and this means that you can access the full source code as well as discuss it, come with suggested improvements, etc.

On the upper right you see the UX of the sample client included in the architecture blueprint, and the functionality is that the combo box is filled with the categories when the application starts. Then, when a category is selected, the product names are shown in the text box below.

Even if this series is complete, we will continue to build further on this architecture blueprint, and any suggestions on things to add are most welcome. Any other feedback, for that matter, is also welcome!

posted on Wednesday, May 28, 2008 6:31:06 AM UTC  by Chris  #    Comments [0]
 Friday, May 23, 2008

WeI will continue the more practical part of this series by showing how the new mobile architecture looks in code, and if you want some background, please see the previ