J2EE Clustering Part 1
By Abraham Kang
Translation Tianyu 2002-12-13
Http://www.javaworld.com/javaworld/jw-02-2001/jw-0223-extremescale_p.html
Flexible telescopic topology
The cluster also requires a flexible layout of a scalable topology. Most application servers can bear a HTTP server is responsibility for an application server. As shown in Figure 1.
Figure 1. All-in-one topology
If your site is dynamic content service, the architecture shown in Figure 1 is good. However, if your site part is a static content service, the scalable site may be an expensive suggestion, because you would rather add more application servers. To serve the static HTML page. Remember, rely on increasing the web server to retract your site static part; rely on the dynamic part of your site to extend your site, as shown in Figure 2.
Figure 2. Partitioned Topology
The main disadvantage of the architecture shown in Figure 2: dynamic page processing adds a reaction time. However, in order to stand independently static and dynamic parts, it provides a flexible approach.
Finally, don't pay attention to maintenance application server discussion??
Maintenance
For a large number of machines in the cluster, maintain the scope of the cluster operation and highlight the application change. Application Server should provide a proxy to aware of the key service when it fails, then restart or activate them on the backup server. Land, when the change or update appears, the application server should provide an easy way to update and synchronize all servers in the cluster.
Sybase Enterprise Application Server and HP Bluestone Total-E-Server provide File and Configuration Synchronization services for clusters. Sybase Enterprise Application Server offers File and Configuration Synchronization services at the hierarchy of the host, packet, or cluster. Bluestone only provides File at the host level. Configuration Synchronization Service. If the big app has many changes to be deployed, this process will take a long time, the BEA WebLogic Server only provides Configuration synchronization. Compare these two, with a storage area network with Configuration synchronization work better Because the change can be placed in a separate logical memory, all machines in the cluster will receive changes to the application file, then each machine must receive a structural configuration from a centralized configuration server. Silverstream Application Server Use a dynamic class loader to load application files and structural configurations from the database. Dynamic class loaders drive application changes in a running application server.
Conclusion is made here for the important feature to consider in the application server. Below, let's take a look at how four popular application servers are compared to our standards.
Application server comparison
We have generally talked about the cluster, now let us pay attention to a single application server, apply what we have learned to the real world. Below, you will find the more:
· HP Bluestone Total-E-Server 7.2.1
· Sybase Enterprise Application Server 3.6
· Silverstream Application Server 3.7
· BEA WebLogic Server 6.0
As in this article, each application server item provides a picture of a HA architecture, which is an outline of an important feature.
HP Bluestone Total-E-Server 7.2.1
Figure 3. HP Bluestone 7.2.1 Topology
Cluster summary:
Bluestone implements a cluster of a separate JNDI tree. In a web server running as a plugin, the LBB runs as a plugin, providing the load balancing and failover of the HTTP request. LBBS knows which application is running on which UBS (Universal Business Server) instance, And it can transmit appropriately. The main disadvantage of STATEFUL and STATELESS SESSION and ENTITELESS SESSION and ENTITELESS SESSION and Entity Beans are supported by EJB proxy services and proxy LBB. The main disadvantage of ENTIB proxy service is to increase the reaction time for each EJB request, and it and it The UBS instance is running on the same machine. In case the UBS instance fails, the EJB proxy service and UBS stubs allow failure, but in case hardware fail, it is not allowed to fail. Configuration or APSERVER.TXT's proxy LBB Configuration support hardware failover. Client APSERVER.TXT lists all components in the cluster. BAM (Bluestone Application Manager) or Host-by-Host Basis, when additional components are added to the cluster You need to update all clients. The APSERVER.TXT isolation client and cluster changes on the agent LBB. But for each EJB call, additional reaction time is added. HP Bluestone is the only application for clustering and load balancing JMS Server. Cluster convergence time (CONVERGENCE TIME):
Comparative comparison of JNDI trees with central concentration and global sharing. Least
HttpSession failed
Status server and backup status server or database in the central concentration of the memory.
Single point failed
None
Flexible cluster topology
Support all cluster topologies.
Maintenance
Bluestone is better than other application servers in maintenance. Bluestone provides a dynamic application distributor (DAL), when an application or machine Down is dropped, the LBB calls it. In addition, Bluestone provides a Bluestone Application Manager (Bam) The configuration and deployment tool. BAM can deploy application packages and their associated configuration files, this tool is unfavorable, you can only configure a host at a time - there is a problem with big clusters.
Sybase Enterprise Application Server 3.6
Figure 4.sybase Enterprise Application Server 3.6 Topology
Cluster summary:
Enterprise Application Server implements the cluster of the central centralized JNDI tree, the hardware load balancer provides the load balancing and failover of HTTP requests. Two named servers for each cluster can handle HTTP requests, but for performance, one of them is Dedicated to JNDI requests alone.
Enterprise Application Server3.6 does not have a web server plug-in; however, they will be available with EBF (Error and Bug fixes) in February. This application supports Stateful and Stateless Session and Entity Beans between method calls. Compliance with Enterprise Application Server does not provide any monitoring agent or dynamic application distributor, requires you to buy a third-party solution like a Veritas cluster server that prevents single point failure or server automatic restart. Enterprise Application Server does not support JMS.
CONVERGENCE TIME:
The convergence time relies on the number of named servers and clusters. In three cluster implementations, the cluster of the central centralized JNDI tree generates the worst time to convergence time. Although convergence Time is important, but member machines After the binding object is used by the naming server used by the EJB client, you can start receiving the request. (Not recommended)
HttpSession failed
HTTPSESSION failed shifts with a centralized database. There is no memory copy option.
Single point failed
If you run multiple named servers, there is no single point failure.
Flexible cluster topology
Due to the shortcomings of the web server plug-in, the flexible cluster topology is restricted maintenance.
Sybase provides the best options for application deployment through a file and configuration synchronization. You can synchronize clusters in components, packages, servlets, applications, or web application hierarchy. You can also choose to synchronize a cluster, machine group, or a single host - - A terrible feature If there are many machines in the cluster, there are many files to increase and update in addition to it. A weak point is a lack of dynamic application sending services, meaning you must purchase third-party solutions such as Veritas cluster server .
Silverstream Application Server 3.7
Figure 5. Silverstream Application Server Dispatcher Topology
Figure 6. Silverstream Application Server Web Server
Integration Module Topology.
Cluster summary:
When setting up a SilverseTream cluster, you should pick two configurations: Based on the distributor or web server integration module (WSI). In the distributor-based cluster, users are connected to the distributor or hardware-based distributors --- For example, Alteon 180E. Then the distributor sends an HTTP redirection to connect the client and the machine in the cluster. As indicated, the client is physically bound to a single server. In the distributor configuration, the single-machine and cluster There is no difference, because it involves the client, the cluster has become a single machine. Main shortcomings: You cannot telescide the static part of the site dynamic part.
In a WSI-based cluster, the user is connected to the distributor, the distributor requests load balancing and failure to the web server. Each web server has a plugin that points to the Silverstream cluster. WSI cluster does not use heavy Direction, so each HTTP request is loaded by any web server. The second load balancer ensures failover in the application server hierarchy. An advantage of establishing a WSI architecture on the distributor architecture: Has a stand-alone telescopic site static And dynamic partial ability. Argopersistencemanager is required for HttpSession failover. In ejB client, the EJB client does not fail between the method call. EjB failed Transfer is entirely the responsibility of the developer. Finally, SilversTream does not support JMS of the cluster.
HttpSession failed
SilversTream relies on the centralized database and the ArgopersistenceManager provides httpsession failover. Unfortunately, this solution is all right. In order to replace the standard HttpSession storage session information to the database, you must use this product's ArgopersistenceManager class - this is a Disadvantages.
CONVERGENCE TIME:
Compared with the central concentration and global sharing JNDI trees. Least.
Single point failed:
Does not exist in the above architecture.
Flexible cluster topology
Support all cluster topologies.
Maintenance:
Cache Manager and Dynamic Loader provide an easy way - using an application that is deployed and updated with uninterrupted activation clients. When updating the app on a server in the cluster, update the logging database simultaneous cache management The cache on all other machines is invalid, enforce their next access to re-apply the update section. This way is the time required to load applications and access to active memory during initialization, and enters the active memory.
BEA WebLogic Server 6.0
Figure 7. BEA WebLogic Server 6.0 Topology.
Cluster summary:
WebLogic Server implements a cluster through the global sharing JNDI cluster tree. By setting, when a single server is started, it puts the JNDI tree to the global shared jndi tree. Then the server uses the copy to the request service, allowing a single server to know all objects in the cluster. Accurate position. The client use Stubs is awareness cluster. It means that it is to say that they are optimized on their original servers, but also understand the situation of all other replica objects running in different application servers. Because the clustered stubs understand the cluster All replicass, they can significantly fail in the cluster to other replica objects. One feature of WebLogic Server is that Stateful session beans and EJB Remotes object configurable automatic failover transfers are memory copies .WebLogic definition clusterable as running A service in the cluster. JMS is ClusteRable but each Topic or Queue runs only on a separate server in the cluster. You cannot load balanced or failover JMS Topics or queues in the cluster - this is WebLogic JMS implementation A major weakness. Httpsession failed
WebLogic Server relies on a clear memory state to copy to an arbitrary backup server or database server to provide HttpSession failure. Each machine in the cluster selects a different backup machine. If the primary server fails, backups become the first machine At the same time, this primary machine picks up new backup servers. One feature ofWeblogic is the independent cookie. HP Blueestone and Enterprise Application Server requires cookies for httpsession failover. If there is a failure, WebLogic can use encrypted information in the URL Position the client to the backup server.
Single point failed:
JMS and Administration Server.
Flexible cluster topology
Support all cluster topologies.
Maintenance:
WebLogic's weakness is maintenance. Although BEA has taken positive measures about configuration. Theweblogic server does not provide monitoring agents, dynamic application distributors, or file synchronization services. These shortcomings need you to buy third parties about solving single point failure or implementation Potential HA solution, all WebLogic administrators must put on Nike Air Streak Vapor IV Racing Shoes. If a server Down has dropped, they must run to the problem of prior to machinery. However, if you use SAN, you will No file synchronization service is required. However, most developers have just begun to recognize the benefits of SAN.
analysis
In general, the BEA WebLogic Server 6.0 has the cluster implementation after the most robust and cautious consideration. HP Bluestone Total-E-Server 2.7.1, Sybase Enterprise Application Server 3.6, and Silverstream Application Server 3.7 is usually the next choice .
Select the appropriate application server needs to be trade-off. If you are writing an app, you will have an eJB Clients (Applets and Applications), the extended use of the web clients, httpsession (for caching), easy to telescope and fail. You will BEA Weblogic 6.0 will be used. If your application needs a lot of JMS and your client is mostly a web client, Bluestone may be a better choice. From the viewpoint of the cluster, Sybase Enterprise Application Server 3.7 lacks important Cluster features, such as JMS, memory httpsession replication, server affinity, server affinity, and split through the web server plugin. However, Sybase Enterprise Application Server 3.7 brings Table File and Structure Configuration Synchronization Services. But if you With SAN, the benefits are minimized. Silverstream Application Server has the weakest cluster implementation, lacks JMS of the cluster, without memory httpsession copy and fail transfer.
in conclusion
In this article, you have already had a basic understanding of clusters, cluster methods, and important clusters. We have studied the benefits and disadvantages of each application server, and discuss the important features of finding related to clusters in an application server. To finish these knowledge, now you understand how to set up high availability and scalability. However, these descriptions are just your trip, go out to get some cluster licenses evaluation, Control-C Some application servers. We will start in part2. Write code and test the application server provider claims and provides failed transfer and cluster policies for each different application server. For load testing and cluster convergence benchmarks, we will use J2EE's Java PetStore.
About author
Abraham Kang is a scalability and high availability architect of Infogain's Delivery Operations Group. He has a year and a half J2EE application server cluster experience. In addition, he is still a SUN certified programmer and developer, Oracle 8i professional certification DBA, CCNA, and CCDA Abraham wanted to thank him, Jessie Spencer-Cooke, gave him an excellent environment and supported him to work hard.
-------------------------------------------------- -----------------------------
HA material list
Network infrastructure
In a HA's intranet, you should have all the content below:
· Firewall with redundant and failure transfer capabilities
· Redundant gateway router running on the thermal spare Routing Protocol (HSRP)
· Redundant connection to different ISPs, run the Boundary Gateway Protocol (BGP) with different Internet link centers (such as AT & T and Sprint)
· Redundant SAN swap infrastructure
· Load balancer / distributor with redundant and fail transfer capabilities
• Run the spanning tree protocol (STP) instead of the redundant switching network for shared media networks.
• Redundant SAN Dual Switching is connected to the Dual Ported disk drive using XOR Logic (RAID5) in a dual cycle disk array.
Database infrastructure
In an HA's database environment, you should have a hot backup database.
Storage infrastructure
Storage Area Network (SANS) represents a network technology of high-speed connection servers and memory (disks) in recent years. SANs uses large-capacity storage infrastructure to connect the LAN network model to provide failed transfer, fail return and scalability. Put SAN See the logical-driven client, read and write data through the SAN switch to the network. The SAN switch has the function of managing the actual data to write. For failure transfer, the HA SAN is connected to the mirror disk array. If A disk array Down is dropped, backup immediately take over without affecting the client. SANS rely on sends a replacement disk controller to provide failed return, the merged disk controller enters SAN, so it can come out I / O. In addition, SANs allow you In SAN rely on multiple copied explicit returning your memory needs to eliminate some bottlenecks.
