简单LRU算法实现缓存-update2

update1:第二个实现,读操作不必要采用独占锁,缓存显然是读多于写,读的时候一开始用独占锁是考虑到要递增计数和更新时间戳要加锁,不过这两个变量都是采用原子变量,因此也不必采用独占锁,修改为读写锁。
update2:一个错误,老是写错关键字啊, LRUCache的 maxCapacity应该声明为volatile,而不是transient。
  
   最简单的LRU算法实现,就是利用jdk的LinkedHashMap,覆写其中的removeEldestEntry(Map.Entry)方法即可,如下所示:
import  java.util.ArrayList;
import  java.util.Collection;
import  java.util.LinkedHashMap;
import  java.util.concurrent.locks.Lock;
import  java.util.concurrent.locks.ReentrantLock;
import  java.util.Map;


/**
 * 类说明:利用LinkedHashMap实现简单的缓存, 必须实现removeEldestEntry方法,具体参见JDK文档
 * 
 * 
@author  dennis
 * 
 * 
@param  <K>
 * 
@param  <V>
 
*/
public   class  LRULinkedHashMap < K, V >   extends  LinkedHashMap < K, V >  {
    
private   final   int  maxCapacity;

    
private   static   final   float  DEFAULT_LOAD_FACTOR  =   0.75f ;

    
private   final  Lock lock  =   new  ReentrantLock();

    
public  LRULinkedHashMap( int  maxCapacity) {
        
super (maxCapacity, DEFAULT_LOAD_FACTOR,  true );
        
this .maxCapacity  =  maxCapacity;
    }

    @Override
    
protected   boolean  removeEldestEntry(java.util.Map.Entry < K, V >  eldest) {
        
return  size()  >  maxCapacity;
    }
    @Override
    
public   boolean  containsKey(Object key) {
        
try  {
            lock.lock();
            
return   super .containsKey(key);
        } 
finally  {
            lock.unlock();
        }
    }

    
    @Override
    
public  V get(Object key) {
        
try  {
            lock.lock();
            
return   super .get(key);
        } 
finally  {
            lock.unlock();
        }
    }

    @Override
    
public  V put(K key, V value) {
        
try  {
            lock.lock();
            
return   super .put(key, value);
        } 
finally  {
            lock.unlock();
        }
    }

    
public   int  size() {
        
try  {
            lock.lock();
            
return   super .size();
        } 
finally  {
            lock.unlock();
        }
    }

    
public   void  clear() {
        
try  {
            lock.lock();
            
super .clear();
        } 
finally  {
            lock.unlock();
        }
    }

    
public  Collection < Map.Entry < K, V >>  getAll() {
        
try  {
            lock.lock();
            
return   new  ArrayList < Map.Entry < K, V >> ( super .entrySet());
        } 
finally  {
            lock.unlock();
        }
    }
}
    如果你去看LinkedHashMap的源码可知,LRU算法是通过双向链表来实现,当某个位置被命中,通过调整链表的指向将该位置调整到头位置,新加入的内容直接放在链表头,如此一来,最近被命中的内容就向链表头移动,需要替换时,链表最后的位置就是最近最少使用的位置。
    LRU算法还可以通过计数来实现,缓存存储的位置附带一个计数器,当命中时将计数器加1,替换时就查找计数最小的位置并替换,结合访问时间戳来实现。这种算法比较适合缓存数据量较小的场景,显然,遍历查找计数最小位置的时间复杂度为O(n)。我实现了一个,结合了访问时间戳,当最小计数大于MINI_ACESS时(这个参数的调整对命中率有较大影响),就移除最久没有被访问的项:
package  net.rubyeye.codelib.util.concurrency.cache;

import  java.io.Serializable;
import  java.util.ArrayList;
import  java.util.Collection;
import  java.util.HashMap;
import  java.util.Iterator;
import  java.util.Map;
import  java.util.Set;
import  java.util.concurrent.atomic.AtomicInteger;
import  java.util.concurrent.atomic.AtomicLong;
import  java.util.concurrent.locks.Lock;
import  java.util.concurrent.locks.ReadWriteLock;
import  java.util.concurrent.locks.ReentrantLock;
import  java.util.concurrent.locks.ReentrantReadWriteLock;

/**
 * 
 * 
@author  dennis 类说明:当缓存数目不多时,才用缓存计数的传统LRU算法
 * 
@param  <K>
 * 
@param  <V>
 
*/
public   class  LRUCache < K, V >   implements  Serializable {

    
private   static   final   int  DEFAULT_CAPACITY  =   100 ;

    
protected  Map < K, ValueEntry >  map;

    
private   final  ReadWriteLock lock  =   new  ReentrantReadWriteLock();

    
private   final  Lock readLock  =  lock.readLock();

    
private   final  Lock writeLock  =  lock.writeLock();

    
private   final volatile int  maxCapacity;  //保持可见性

    
public   static   int  MINI_ACCESS  =   5 ;

    
public  LRUCache() {
        
this (DEFAULT_CAPACITY);
    }

    
public  LRUCache( int  capacity) {
        
if  (capacity  <=   0 )
            
throw   new  RuntimeException( " 缓存容量不得小于0 " );
        
this .maxCapacity  =  capacity;
        
this .map  =   new  HashMap < K, ValueEntry > (maxCapacity);
    }

    
public   boolean  ContainsKey(K key) {
        
try  {
            readLock.lock();
            
return   this .map.containsKey(key);
        } 
finally  {
            readLock.unlock();
        }
    }

    
public  V put(K key, V value) {
        
try  {
            writeLock.lock();
            
if  ((map.size()  >  maxCapacity  -   1 &&   ! map.containsKey(key)) {
                
//  System.out.println("开始");
                Set < Map.Entry < K, ValueEntry >>  entries  =   this .map.entrySet();
                removeRencentlyLeastAccess(entries);
            }
            ValueEntry new_value 
=   new  ValueEntry(value);
            ValueEntry old_value 
=  map.put(key, new_value);
            
if  (old_value  !=   null ) {
                new_value.count 
=  old_value.count;
                
return  old_value.value;
            } 
else
                
return   null ;
        } 
finally  {
            writeLock.unlock();
        }
    }

    
/**
     * 移除最近最少访问
     
*/
    
protected   void  removeRencentlyLeastAccess(
            Set
< Map.Entry < K, ValueEntry >>  entries) {
        
//  最小使用次数
         long  least  =   0 ;
        
//  访问时间最早
         long  earliest  =   0 ;
        K toBeRemovedByCount 
=   null ;
        K toBeRemovedByTime 
=   null ;
        Iterator
< Map.Entry < K, ValueEntry >>  it  =  entries.iterator();
        
if  (it.hasNext()) {
            Map.Entry
< K, ValueEntry >  valueEntry  =  it.next();
            least 
=  valueEntry.getValue().count.get();
            toBeRemovedByCount 
=  valueEntry.getKey();
            earliest 
=  valueEntry.getValue().lastAccess.get();
            toBeRemovedByTime 
=  valueEntry.getKey();
        }
        
while  (it.hasNext()) {
            Map.Entry
< K, ValueEntry >  valueEntry  =  it.next();
            
if  (valueEntry.getValue().count.get()  <  least) {
                least 
=  valueEntry.getValue().count.get();
                toBeRemovedByCount 
=  valueEntry.getKey();
            }
            
if  (valueEntry.getValue().lastAccess.get()  <  earliest) {
                earliest 
=  valueEntry.getValue().count.get();
                toBeRemovedByTime 
=  valueEntry.getKey();
            }
        }
        
//  System.out.println("remove:" + toBeRemoved);
        
//  如果最少使用次数大于MINI_ACCESS,那么移除访问时间最早的项(也就是最久没有被访问的项)
         if  (least  >  MINI_ACCESS) {
            map.remove(toBeRemovedByTime);
        } 
else  {
            map.remove(toBeRemovedByCount);
        }
    }

    
public  V get(K key) {
        
try  {
            readLock.lock();
            V value 
=   null ;
            ValueEntry valueEntry 
=  map.get(key);
            
if  (valueEntry  !=   null ) {
                
//  更新访问时间戳
                valueEntry.updateLastAccess();
                
//  更新访问次数
                valueEntry.count.incrementAndGet();
                value 
=  valueEntry.value;
            }
            
return  value;
        } 
finally  {
            readLock.unlock();
        }
    }

    
public   void  clear() {
        
try  {
            writeLock.lock();
            map.clear();
        } 
finally  {
            writeLock.unlock();
        }
    }

    
public   int  size() {
        
try  {
            readLock.lock();
            
return  map.size();
        } 
finally  {
            readLock.unlock();
        }
    }

    
public   long  getCount(K key) {
        
try  {
            readLock.lock();
            ValueEntry valueEntry 
=  map.get(key);
            
if  (valueEntry  !=   null ) {
                
return  valueEntry.count.get();
            }
            
return   0 ;
        } 
finally  {
            readLock.unlock();
        }
    }

    
public  Collection < Map.Entry < K, V >>  getAll() {
        
try  {
            readLock.lock();
            Set
< K >  keys  =  map.keySet();
            Map
< K, V >  tmp  =   new  HashMap < K, V > ();
            
for  (K key : keys) {
                tmp.put(key, map.get(key).value);
            }
            
return   new  ArrayList < Map.Entry < K, V >> (tmp.entrySet());
        } 
finally  {
            readLock.unlock();
        }
    }

    
class  ValueEntry  implements  Serializable {
        
private  V value;

        
private  AtomicLong count;

        
private  AtomicLong lastAccess;

        
public  ValueEntry(V value) {
            
this .value  =  value;
            
this .count  =   new  AtomicLong( 0 );
            lastAccess 
=   new  AtomicLong(System.nanoTime());
        }

        
public   void  updateLastAccess() {
            
this .lastAccess.set(System.nanoTime());
        }

    }
}



你可能感兴趣的:(简单LRU算法实现缓存-update2)