日本亚洲中午字幕乱码,亚洲乱码在线播放,亚洲午夜精品久久久久久浪潮

JGroup UDP和TCP两种方式

w@ns0ng — Thu, 25 Aug 2011 00:18:00 GMT

JGroup可以��Z��TCP协议来实现消息广播，也可以通过UDP方式来广播消息，利弊不言而喻�Q�TCP可靠�Q�但是代价大�Q�性能没有UDP来的好，UDP速度快，代�h��，但是消息的丢��q��以及无序性有着很大的限制。但是JGroup在UDP方式的基��上，增加了协议栈的配�|�，通过配置上层的协议，可以保证消息的重发，大包体的分解�Q�同时保证消息包体顺序）�Q�组内机器的状态检��等功能�?br />

http://www.javachen.com/2011/06/jgroups-introduction-and-configruation/

http://blog.csdn.net/lnfszl/article/details/5747427

http://docs.jboss.org/jbossas/jboss4guide/r4/html/jbosscache.chapt.html

w@ns0ng 2011-08-25 08:18 发表评论

w@ns0ng — Sat, 06 Aug 2011 12:19:00 GMT

目前��目中在克服JGroups初期使用的困难之后，已经使用比较�E�_��了。感觉比较烦琐和�Ҏ��出错的还是JGroups配置。感觉中文的资源较少�Q�现��自己�ȝ��的经验�ȝ��如下
Tim http://hi.baidu.com/jabber/blog/item/7e879852a23efd0f0cf3e3ea.html

http://puras.iteye.com/blog/81783

JGroups 适合使用场合
服务器集��cluster、多服务器通讯、服务器replication(复制)�{�，分布式cache�~�存

JGroups ��?/strong>
JGroups是一个基于Java语言的提供可靠多�?�l�播)的开发工具包。在IP Multicast基础上提供可靠服务，也可以构建在TCP或者WAN上。主要是由Bela Ban开发，属于JBoss.org�Q�在JBoss的网站也有一些相��x��。目前在 SourceForge上还是比较活跃，�l�常保持更新�?br />
JGroups 配置

PING: 发现初始成员
MERGE2: ��网�l�层切分的包重新合�ƈ�?br />FD_SOCK: Failure Dectection 错误��，��Z��TCP
FD�Q�Failure Dectection 错误��，��Z��心蟩
VERIFY_SUSPECT: ��查貌似失败的节点
pbcast.NAKACK: 应答�Q�提供可靠传�?br />UNICAST: 可靠的UNICAST
pbcast.STABLE: 计算�q�播信息是否�E�_��
VIEW_SYNC: 定期�q�播view(成员名单)
pbcast.GMS: Group membership, 处理joins/leaves/crashes�{?br />FC: ��量控制
FRAG2:Fragmentation layer�Q�分包，��大的数据包分拆成适合�|�络层传�?br />
以上一些是比较重要的配�|�，基本上不能少。如果要深入研究可以在　org.jgroups.protocols　里面查看源代�?br />
JGroups使用例子, JGroups demo, Tim的hello world例子
Timreceiver.java

import org.jgroups.tests.perf.Receiver;
import org.jgroups.tests.perf.Transport;
import org.jgroups.util.Util;

public class TimReceiver implements Receiver {
private Transport transport = null;

public static void main(String[] args) {
TimReceiver t = new TimReceiver();
try {
int sendMsgCount = 5000;
int msgSize = 1000;
t.start();

t.sendMessages(sendMsgCount, msgSize);
System.out.println("########## Begin to recv...");
Thread.currentThread().join();
} catch (Exception e) {
e.printStackTrace();
} finally {
if (t != null) {
t.stop();
}
}
}

public void start()
throws Exception {
transport = (Transport) new TimTransport();
transport.create(null);
transport.setReceiver(this);
transport.start();
}

public void stop() {
if (transport != null) {
transport.stop();
transport.destroy();
}
}

private int count = 0;
public void receive(Object sender, byte[] data) {
System.out.print(".");
if (++count == 5000) {
System.out.println("\r\nRECV DONE.");
System.exit(0);
}

}

private void sendMessages(int count, int msgSize)
throws Exception {
byte[] buf = new byte[msgSize];
for (int k = 0; k < msgSize; k++)
buf[k] = 'T';

System.out.println("-- sending " + count + " " + Util.printBytes(msgSize) + " messages");

for (int i = 0; i < count; i++) {
transport.send(null, buf);
}

System.out.println("######### send complete");
}
}

TimTransport.java

import java.util.Map;
import java.util.Properties;

import org.jgroups.Address;
import org.jgroups.JChannel;
import org.jgroups.Message;
import org.jgroups.ReceiverAdapter;
import org.jgroups.tests.perf.Receiver;
import org.jgroups.tests.perf.Transport;

public class TimTransport extends ReceiverAdapter implements Transport{
private JChannel channel = null;
private String groupName = "TimDemo";
private Receiver receiver = null;

String PROTOCOL_STACK_UDP1 = "UDP(bind_addr=192.168.100.59";
String PROTOCOL_STACK_UDP2 = ";mcast_port=8888";
String PROTOCOL_STACK_UDP3 = ";mcast_addr=225.1.1.1";
String PROTOCOL_STACK_UDP4 = ";tos=8;loopback=false;max_bundle_size=64000;" +
"use_incoming_packet_handler=true;use_outgoing_packet_handler=false;ip_ttl=2;enable_bundling=true):"
+ "PING:MERGE2:FD_SOCK:FD:VERIFY_SUSPECT:"
+"pbcast.NAKACK(gc_lag=50;max_xmit_size=50000;use_mcast_xmit=false;" +
"retransmit_timeout=300,600,1200,2400,4800;discard_delivered_msgs=true):"
+"UNICAST:pbcast.STABLE:VIEW_SYNC:"
+"pbcast.GMS(print_local_addr=false;join_timeout=3000;" +
"join_retry_timeout=2000;" +
"shun=true;view_bundling=true):"
+"FC(max_credits=2000000;min_threshold=0.10):FRAG2(frag_size=50000)";

public Object getLocalAddress() {
return channel != null ? channel.getLocalAddress() : null;
}

public void start() throws Exception {
channel.connect(groupName);
}

public void stop() {
if (channel != null) {
channel.shutdown();
}
}

public void destroy() {
if (channel != null) {
channel.close();
channel = null;
}
}

public void setReceiver(Receiver r) {
this.receiver = r;
}

public Map dumpStats() {
return channel != null ? channel.dumpStats() : null;
}

public void send(Object destination, byte[] payload) throws Exception {
byte[] tmp = new byte[payload.length];
System.arraycopy(payload, 0, tmp, 0, payload.length);
Message msg = null;
msg = new Message((Address) destination, null, tmp);
if (channel != null) {
channel.send(msg);
}
}

public void receive(Message msg) {
Address sender = msg.getSrc();
byte[] payload = msg.getBuffer();
if (receiver != null) {
try {
receiver.receive(sender, payload);
} catch (Throwable tt) {
tt.printStackTrace();
}
}
}

public void create(Properties config) throws Exception {
String PROTOCOL_STACK = PROTOCOL_STACK_UDP1 + PROTOCOL_STACK_UDP2 + PROTOCOL_STACK_UDP3 + PROTOCOL_STACK_UDP4;
channel = new JChannel(PROTOCOL_STACK);
channel.setReceiver(this);
}

public void send(Object destination, byte[] payload, boolean oob) throws Exception {
send(destination, payload);
}
}

w@ns0ng 2011-08-06 20:19 发表评论

w@ns0ng — Sat, 06 Aug 2011 12:00:00 GMT
     摘要: 一、需求前�D�|��间做了一个项目，在后台有很多的数据都攑օ��C��cache中了�Q�而且�q�会对cache中的数据�q�行更新。如果只有一台server没有��M��问题�Q�但是如果考虑到集��负载��^衡，�q�接多个server的时候，��有问题出现了，怎么��h��能保证多个server之间cache的同步呢�Q�请看下面的部��v图�?二、引入JGroupsJGroups是一个可靠的�l�间通讯工具�Q�进�E�可以加入一个通讯�l�，�l�组内所有的成员...  阅读全文

w@ns0ng 2011-08-06 20:00 发表评论

使用开源Grid�q�_��Q�GridGain实现�|�格计算

w@ns0ng — Sat, 06 Aug 2011 06:59:00 GMT

�|�格计算一般分��Z��U�：数据�|�格和计��网��|��单的说数据网格就是把数据分布式存储，计算�|�格��是��Q务分解�ؓ子认为�ƈ行计��?/p>
一个计��网格��^台的作用��是��Q务分解开来，交给不同的结�Ҏ��器运行，然后把运行结果汇聚�v来。这��是Split and Aggregate。如下图所�C�，一个job��h��分解��Z��个sub-job�Q�分别被不同的机器执行，然后把结果汇聚，�q�回�l�调用的客户�?/p>

GridGain是一个开源的java�|�格�q�_��。它集成了很多现成的框架�Q�例�?/p>
JBoss
Spring
Spring AOP
JBoss AOP
AspectJ
JGroups

GridGain有两个方法将应用�E�序grid化：

�W�一�U�是使用AOP

假设有一个应�?/p>
01 class BizLogic {
02   public static Result process(String param) {
03     …
04   }
05 }
06
07 class Caller {
08   public static void Main(String[] args) {
09     BizLogic.process(args[0]);
10   }
11 }

如果要把process grid化，只要��单的使用一个@Gridify的annotation卛_��Q�在Caller客户端要启动GridFactory

01 class BizLogic {
02   @Gridify(…)Here
03   public static Result process(String param) {
04     …
05   }
06 }
07
08 class Caller {
09   public static void Main(String[] args) {
10     GridFactory.start();Here
11
12     try {
13       BizLogic.process(args[0]);
14     }
15     finally {
16       GridFactory.stop();Here
17     }
18   }
19 }

w@ns0ng 2011-08-06 14:59 发表评论

w@ns0ng — Sat, 06 Aug 2011 06:11:00 GMT

转蝲 http://www.iteye.com/topic/475010

�|�格技术分��Z��U�一�U��ؓ�|�格计算一�U��ؓ�|�格数据�Q�gridgain是网��D��，可以�Ҏ��规则分节�Ҏ��术，提高计算速度�Q�比如用于规则引擎，技术引擎等�Q�银行结��，保险考核�l�算�{�。但是如果在计算中数据时分布式或者是延时加蝲�Ӟ��此时数据获取不到�Q�这个时候就需要用到网格数据，在网��g��计算�Q�在�|�格中获取数据，�q�个时�?strong>Infinispan��是一个不错的选择�Q?br />

数据分区技术和数据�|�格的集�?/span>

概述�Q?/span>

当处理大量数据的时候，常常值得推荐的是跨节�Ҏ��数据分隔开处理。基本上�Q�每个点负责处理数据的一部分。这�U�方法基本上允许从数据库数据中加载大量的数据到缓存，然后配置你的电脑区执行这些数据。�ؓ什么？��Z��避免数据在各节点的重复缓�Ԍ��q�样往往可以提升性能�Q�防止服务器瘫痪�?/span>

使用gridgain�Q��?/span>Affinity Load Balancing�q�样的设计非常完��的解决了这个问题，而且可以和分布式�~�存集成�Q�解��x��据网根{�?/span>

Affinity Load Balancing

�?/span>GridGain�?/span>Affinity Load Balancing是通过GridAffinityLoadBalancingSpi.提供�?/span>

下图说明是��用数据网格和不适用数据�|�格的差别。左面的图表�C�没使用GridGain的执行流�E�，其中�q�程数据库服务器负责查询数据�Q�然后传递到主调用服务器。这�U�比数据库访问要快，但是�l�果计算使��用很多不必要的流量�?/span>

叛_��Q?/span>使用�?/span>Gridgain。整个逻辑计算与数据访问整合到本地节点。假讑֤�量逻辑计算比数据序列到数据库要��d�y�Q�即大量计算�Q�，那么�|�络��量��是最��的。此外，您的计算都可以访问节�?/span>2和节�?/span>3的数据。在�q�种情况下，GridGain��分为逻辑计算jobs和合适的逻辑计算路由到相应的数据服务中进行计��。以��保所有计��都在本地节点中计算。现在，如果数据服务节点崩溃�Ӟ��您的��p�|jobs会自动�{�U�d��其他节点�Q�这�U�是允许��p�|的（数据�|�格和分布式�~�存提供�q�种方式�Q��?/span>

数据�|�格集成

GridGain没有实现数据高速缓存，但是与现有的�?/span> 据高速缓存或数据�|�格解决�Ҏ��q�行了集成。这使用户可以��用几乎�Q何的分布式缓存来实现自己喜欢的方案�?/span>

比如�Q?/span>GridGain提供了一�?/span>JBoss Cache Data Partitioning Example 告诉用户如何来��?/span>Attinty Load Balancing。事实上�Q?/span>JBOSSCache没有提供数据分区的功能。由于��用了GridGain�?/span>GridAffinityLoadBalancingSpi提供�?/span>Attinity Load Balancing�?/span>JBoss Cache 数据分区成�ؓ了可能�?br />
本文包含附�g�Q�admin@pjprimer.com 索取
==========English==========
Overview: When processing mass data, what often be worth to recommend is to cross node to open data space processing. Basically, every order the one share of responsible processing data. This kind of method basically allows to arrive from the data with the much to load in database data cache, the computer division that deploys you next carries out these data. Why? To avoid the data repetition in each node amortize, often can promote property so, prevent a server to break down. Use Gridgain, the settlement with such very ideal design of use Affinity Load Balancing this problem, and can mix distributed cache is compositive, solve data reseau. Affinity Load Balancing is in GridGain Affinity Load Balancing is to pass GridAffinityLoadBalancingSpi. Offer. Specification laying a plan is the difference of service data reseau and reseau of not applicable data. The graph of left expresses to did not use the executive technological process of GridGain, among them long-range database server is in charge of inquiring data, deliver next advocate call a server. Than the database the visit wants this kind fast, but as a result computation makes use a lot of needless flow. Right graph, used Gridgain. Whole and logistic computation and data visit conformity arrive this locality node. Assume a large number of logistic computation compare data alignment to want to the database deft (calculate in great quantities namely) , so network discharge will be the smallest. In addition, your computation can visit node 2 with node the data of 3. Below this kind of circumstance, gridGain calculates component for logic Jobs and right logistic computation way by have consideration in corresponding data service. In order to ensure all computation are calculated in this locality node. Now, when if data serves node,breaking down, your unsuccessful Jobs can transfer other node automatically, this kind is allow failure (data reseau and distributed cache offer this kind of way) . [Compositive GridGain did not realize Img][/img] data reseau cache of data high speed, but undertook with cache of existing data high speed or data reseau solution compositive. This makes the user can be used almost any distributed cache will implement the plan that he likes. For instance: GridGain offerred a JBoss Cache Data Partitioning Example to tell an user how to use Attinty Load Balancing. In fact, JBOSSCache did not provide the function of data partition. Because used the Attinity Load Balancing that the GridAffinityLoadBalancingSpi of GridGain offers to let JBoss Eamil ask for: Admin@pjprimer.com

w@ns0ng 2011-08-06 14:11 发表评论

日本亚洲中午字幕乱码,亚洲乱码在线播放,亚洲午夜精品久久久久久浪潮

JGroup UDP和TCP两种方式

使用开源Grid�q�_���Q�GridGain实现�|�格计算

使用开源Grid�q�_��Q�GridGain实现�|�格计算