Posted by: Iqbal M. Khan;

Alachisoft; www.alachisoft.com

Abstract

Caching greatly improves application performance because it reduces expensive trips to the database. But, if you want to use caching in your application, you must decide what to cache and where to put your caching code. The answer is simple. Cache your domain objects and put caching code inside your persistence classes. Domain objects are central to any application and represent its core data and business validation rules. And, while domain objects may keep some read-only data, most of the data is transactional and changes frequently. Therefore, you cannot simply keep domain objects as "global variables" for the entirety of your application because the data will change in the database and your domain objects will become stale, thereby causing data integrity problems. You'll have to use a proper caching solution for this. And, your options are ASP.NET Cache, Caching Application Block in Microsoft Enterprise Library, or some commercial solution like NCache from Alachisoft. Personally, I would advise against using ASP.NET Cache since it forces you to cache from presentation layer (ASP.NET pages) which is bad. The best place to embed caching in your application is your domain objects persistence classes. In this article, I am extending an earlier design pattern I wrote called Domain Objects Persistence Pattern for .NET. I am going to show you how you can incorporate intelligent caching into your application to boost its performance and what considerations you should keep in mind while doing that.

Domain Objects Caching Pattern attempts to provide a solution for domain object caching. The domain objects in this pattern are unaware of the classes that persist them or whether they're being cached or not, because the dependency is only one-way. This makes the domain object design much simpler and easier to understand. It also hides the caching and persistence code from other subsystems that are using the domain objects. This also works in distributed systems where only the domain objects are passed around.

Scope

Domain Objects, Domain Objects Caching.

Problem Definition

Domain objects form the backbone of any application. They capture data model from the database and also the business rules that apply to this data. It is very typical for most subsystems of an application to rely on these common domain objects. And, usually applications spend most of their time in either loading or saving these domain objects to the database. The actual "processing time" of these objects is very small specially for N-Tier applications where each "user request" is very short.

This means that performance of the application depends greatly on how quickly these domain objects can be made available to the application. If the application has to make numerous database trips, the performance is usually bad. But, if the application caches these objects close-by, the performance improves greatly.

At the same time, it is very important that domain object caching code is kept in such a central place that no matter who loads or saves the domain objects, the application automatically interacts with the cache. Additionally, we must hide the caching code from the rest of application so we can take it out easily if needed.

Solution

As described above, the solution is an extension of an existing design pattern called Domain Objects Persistence Pattern for .NET. That pattern already achieves the goal of separating domain objects from persistence code and from the rest of the application as well. This double-decoupling provides a great deal of flexibility in the design. The domain objects and the rest of the application is totally unaffected whether the data is coming from a relational database or any other source (e.g. XML, flat files, or Active Directory/LDAP).

Therefore, the best place to embed caching code is in the persistence classes. This ensures that no matter which part of the application issues the load or save call to domain objects, caching is appropriately referenced first. This also hides all the caching code from rest of the application and lets you replace it with something else should you choose to do so.

Domain and Persistence Classes

In this sample, we will look at an Employee class from Northwind database mapped to the "Employees" table in the database.

// Domain object "Employee" that holds your data public class Employee {     // Some of the private data members     // ...     public Employee() {}     // Properties for Employee object     public String    EmployeeId { get {return _employeeId;} set {_employeeId = value;}}     public String    Title { get {return _title;} set {_title = value;}}     public ArrayList Subordinates { get {return _subordinates;} set {_subordinates = value;}} } // Interface for the Employee persistence public interface IEmployeeFactory  {     // Standard transactional methods for single-row operations     void Load(Employee emp);        void Insert(Employee emp);      void Update(Employee emp);      void Delete(Employee emp);      // Load the related Employees (Subordinates) for this Employee     void LoadSubordinates(Employee emp);     // Query method to return a collection of Employee objects     ArrayList FindByTitle(String title); } // Implementation of Employee persistence public class EmployeeFactory : IEmployeeFactory {     // all methods described in interface above are implemented here } // A FactoryProvider to hide persistence implementation public class FactoryProvider  {     // To abstract away the actual factory implementation     public static IEmployeeFactory GetEmployeeFactory() { return new EmployeeFactory(); } }

Sample Application

Below is an example of how a client application will use this code.

public class NorthwindApp {     static void Main (string[] args) {         Employee emp = new Employee();         IEmployeeFactory iEmpFactory = FactoryProvider.GetEmployeeFactory();         // Let's load an employee from Northwind database.         emp.EmployeeId = 2;         iEmpFactory.load(emp);         // Pass on the Employee object         HandleEmployee(emp);         HandleSubordinates(emp.Subordinates);         // empList is a collection of Employee objects                      ArrayList empList = iEmpFactory.FindByTitle("Manager");      } }

The code above shows you the overall structure of your classes for handling domain objects persistence and caching. As you can see, there is clear-cut separation between the domain and persistence classes. And, there is an additional FactoryProvider class that lets you hide the persistence implementation from rest of the application. However, the domain objects (Employee in this case) moves around throughout the application.

Creating Cache Keys

Most cache systems provide you with a string-based key. At the same time, the data that you cache consists of various different classes ("Customers", "Employees", "Orders", etc.). In this situation, an EmployeeId of 1000 may conflict with an OrderId of 1000 if you keys do not contain any  type information. Therefore, you need to store some type information as part of the key as well. Below are some suggested key structures. You can make up your own based on the same principles.

1. Keys for individual objects: If you're only storing individual objects, you can make up your keys as following:
   1. "Customers:PK:1000". This means Customers object with primary key of 1000.
2. Keys for related objects: For each individual object, you may also want to keep related objects so you can easily find them. Here are keys for that:
   1. "Customers:PK:1000:REL:Orders". This means an Orders collection for Customer with primary key of 1000
3. Keys for query results: Sometime, you run queries that return a collection of objects. And, these queries may also take different run-time parameters each time. You want to store these query results so the next time you don't have to run the query. Here are the keys for that. Please note that these keys also include run-time parameter values:
   1. "Employees:QRY:FindByTitleAndAge:Manager:40". This represents a query in "Employees" class called "FindByTitleAndAge" which takes two run-time parameters. The first parameter is "Title" and second is "Age". And, their runtime parameter values are specified.

Caching in Transactional Operations

Most transactional data contains single-row operations (load, insert, update, and delete). These methods are all based on primary key values of the object and are the ideal place to start putting caching code. Here is how to handle each method:

1. Load Method: First check the cache. If data found, get it from there. Otherwise, load from the database and then put in the cache.
2. Insert Method: After successfully adding a row in the database, add its object to the cache as well.
3. Update Method: After successfully updating a row in the database, update its object in the cache as well.
4. Delete Method: After successfully removing a row from the database, remove its object from the cache as well.

Below is a sample Load method with caching logic included. Remember, you're only loading a single object (single row) from the database.

// Check the cache before going to the database void Load(Employee emp) {     try     {         // Construct a cache-key to lookup in the cache first         // The cache-key for the object will be like this: Employees:PK:1000         string objKey = CacheUtil.GetObjectKey("Employee", emp.EmployeeId.ToString());         object obj = CacheUtil.Load(objKey);         if (obj == null)         {             // item not found in the cache. Load from database and then store in the cache             _LoadFromDb(emp);             // For simplicity, let's assume this object does not depend on anything else             ArrayList dependencyKeys = null;             CacheItemRemovedCallback onRemoved = null;             CacheUtil.Store(objKey, emp, dependencyKeys, Cache.NoAbsoluteExpiration,                             Cache.NoSlidingExpiration, CacheItemPriority.Default, onRemoved );             // Now, load all its related subordinates             LoadSubordinates(emp);         }         else         {             emp.Copy((Employee)obj);         }     }     catch (Exception e)     {         // Handle exceptions here     } }

Please note a few things here.

1. RemovedItemCallback: This is a delegate that allows your application to be notified asynchronously when the given item is removed from the cache.
2. Expiration: You can specify an absolute time or idle time expiration. Although, we did not specify any expiration above, you could have specified two types of expirations. One is a fixed-time expiration (e.g. 10 minutes from now) and the second is an idle-time expiration (e.g. if item is idle for 2 minutes).

Caching Relationships

Domain objects usually represent relational data coming from a relational database. Therefore, when you cache them, you have to keep in mind their relationships and cache the related objects as well. And, you also have to create "dependency" between the object and all its related objects. The reason being that if you remove the object from the cache, you should also remove all its related objects so there is not data integrity problems. Below is a code example of how to specify relationships in the cache.

// LoadSubordinates method void LoadSubordinates(Employee emp) {     try     {         // Construct a cache-key to lookup related items in the cache first         // The cache-key for related collection will be like this: Employees:PK:1000:REL:Subordinates         string relKey = CacheUtil.GetRelationKey("Employees", "Subordinates", emp.EmployeeId.ToString());         string employeeKey = CacheUtil.GetObjectKey("Employee", emp.EmployeeId.ToString());         object obj = CacheUtil.Load(relKey);         if (obj == null)         {             // Subordinates not found in the cache. Load from database and then store in the cache             _LoadSubordinatesFromDb(emp);             ArrayList subordList = emp.Subordinates;             // Result is a collection of Employee. Let's store each Employee separately in             // the cache and then store the collection also but with a dependency on all the             // individual Employee objects. Then, if any Employee is removed, the collection will also be             // Count + 1 is so we can also put a dependency on the Supervisor             ArrayList dependencyKeys = new ArrayList(subordList.Count + 1);             for (int index = 0; index , subordList.Count; index++)             {                 string objKey=CacheUtil.GetObjectKey("Employee",subordList[index].EmployeeId.ToString());                 CacheUtil.Store(objKey, subordList[index], null, Cache.NoAbsoluteExpiration,                                         Cache.NoSlidingExpiration, CacheItemPriority.Default, null );                 dependencyKeys[index] = objKey;             }             dependencyKeys[subordList.Count] = employeeKey;             CacheItemRemovedCallback onRemoved = null;             CacheUtil.Store(relKey, subordinateList, dependencyKeys, Cache.NoAbsoluteExpiration,                             Cache.NoSlidingExpiration, CacheItemPriority.Default, onRemoved );         }         else         {             // Subordinates already in the cache. Let's get them             emp.Subordinates = (ArrayList)obj;         }     }     catch (Exception e)     {         // Handle exceptions here     } }

In the above example, you'll notice that a collection is being returned from the database and each object inside the collection is stored individually in the cache. Then, the collection is being cached as a single-item but with a cache dependency on all the individual objects in the collection. This means that if any the individual objects is updated or removed in the cache, the collection is automatically removed by the cache. This allows you to maintain data integrity in caching collections.

You'll also notice in the above example that the collection has a cache dependency on the "primary object" whose related objects the collection contains. This dependency also means that if the primary object is removed or updated in the cache, the collection will be removed in order to maintain data integrity.

Caching in Query Methods

A query method returns a collection of objects based on the search criteria specified in it. It may or may not take any runtime parameters. In our example, we have a FindByTitle that takes "title" as a parameter. Below is an example of how caching is embedded inside a query method.

// Query method to return a collection ArrayList FindByTitle(String title) {     try     {         // Construct a cache-key to lookup items in the cache first         // The cache-key for the query will be like this: Employees:PK:1000:QRY:FindByTitle:Manager         string queryKey = CacheUtil.GetQueryKey("Employees", "Query", title);         object obj = CacheUtil.Load(queryKey);         if (obj == null)         {            // No items found in the cache. Load from database and then store in the cache            ArrayList empList = _FindByTitleFromDb(title);                            // Result is a collection of Employee. Let's store each Employee separately in            // the cache and then store the collection also but with a dependency on all the            // individual Employee objects. Then, if any Employee is removed, the collection will also be            ArrayList dependencyKeys = new ArrayList(empList.Count);            for (int index = 0; index , empList.Count; index++)            {                string objKey = CacheUtil.GetObjectKey("Employee", empList[index].EmployeeId.ToString());                CacheUtil.Store(objKey, empList[index], null, Cache.NoAbsoluteExpiration,                                        Cache.NoSlidingExpiration, CacheItemPriority.Default, null );                dependencyKeys[index] = objKey;            }            CacheItemRemovedCallback onRemoved = null;            CacheUtil.Store(queryKey, empList, dependencyKeys, Cache.NoAbsoluteExpiration,                                 Cache.NoSlidingExpiration, CacheItemPriority.Default, onRemoved );         }         else         {            // Query results already in the cache. Let's get them            return (ArrayList) obj;         }     }     catch (Exception e)     {         // Handle exceptions here     } }

In the above example, just like the relationship method, you'll notice that a collection is being returned from the database and each object inside the collection is stored individually in the cache. Then, the collection is being cached as a single-item but with a cache dependency on all the individual objects in the collection. This means that if any the individual objects is updated or removed in the cache, the collection is automatically removed by the cache. This allows you to maintain data integrity in caching collections.

Applications in Server Farms

The above pattern works for both single-server or server-farm deployment environments. The only thing that must change is the underlying caching solution. Most caching solutions are for single-server environments (e.g. ASP.NET Cache and Caching Application Block). But, there are some commercial solutions like Alachisoft NCache (http://www.alachisoft.com) that provide you a distributed cache that works in a server farm configuration. This way, your application can use a cache from any server in the farm and all cache updates are immediately propagated to the entire server farm.

Conclusion

Using the Domain Objects Caching Pattern, we have demonstrated how you should embed caching code into your persistence classes. And, we've covered the most commonly used situations of Load, Queries, and Relationships with respect to caching. This should give you a good starting point to determine how you should use caching in your application.

Author: Iqbal M. Khan works for Alachisoft, a leading software company providing O/R Mapping and Clustered Object Caching solutions for .NET. You can reach him at iqbal@alachisoft.com or visit Alachisoft at www.alachisoft.com.