Application and database development cluster do you still need ha software?

Application and database development Cluster also need HA software?

The so-called cluster technology (Cluster), also translated into clusters, covering three technical directions: Highly AvailableHA(High availability); Load BalancingLB(Load Balance); High Performance ComputingHPC(High performance Computing). For the business sectorITapplications are primarily concerned with high availabilityHAand load BalancingLBthese two parts. In order to ensure high availability of the application, the operating systemOSthe above developed a number of system-levelHAsoftware. Different operating systems have differentHAsoftware, such asIBM AIXunder the operating systemPowerHA(HACMP),HP-UXunder theMC-SG,Linuxunder theVCS(Veritas Cluster Server),Rose,Heartbeatand so on. There are also some cross-platformHAsoftware, such asIBM TSAsupport multi-platform,HACMPversion5time out.For Linux on POWERthe version (V6after it is gone),MC-SGhave aFor Linuxthe version,VCSThere are also versions of different platforms.

with the development of application software and database software, cluster technology (Clusterare increasingly incorporated into application software and database software. AsOracle RAC, was (Websphere application Server), Weblogic,Tomcatand all the support .Clustertechnology. TheseClustertechnology not only covers load balancingLB (Load Balance)Technology, while also providing highly availableHA(High availability) function. In contrast, the traditional high-availabilityHAsystem software is mainly for the high-availability of this part of the function to provide support. AsPowerHA(HACMP),MC-SG,VCSand so on. TheseHAAlthough the software is available in the main (Primary-backup) feature also provides concurrency (Concurrent) functionality, but its concurrency capability simply provides storage management for concurrent access. In terms of data consistency, applications or database software are still required to reconcile concurrent access and avoid conflicts. In other words, whether concurrency can be achieved depends primarily on the database or application softwareLBfunction. Well, now we need these separateHAsoftware?

first, we want to see that the database and application software itself contains Cluster there is still a limit to the functionality. There are two main limitations: Price variance and functional perfection.

The so-called price difference, refers to different functions of different prices, cluster version of the software will generally have a higher price. InOracle RACFor example, the price is much higher than the standalone version. Of course, compared to the standalone version,RACnot only provides aHAfeatures, while also providing load balancingLBthe function. However, it is important to note that the effect of load balancing is not easy to debug. This is very much related to the complexity of applying access data. For a database,RACjust increased theCPUand theMemory, theDisk IOhas not increased. The limiting factors for system performance are mainlyDisk IOThe application scenario,RACThere is no performance boost. In addition, becauseOraclein-memorySGAin the district willCachelarge amounts of data for application read-write access,RACunder theCache Fusionmechanism is to combine multipleRACnode'sSGAwork, the district must beRACtransmission of synchronized data in the heartbeat network, which causes a largeOverhead(additional overhead). Therefore, if you expect to passRACMulti-node concurrency to improve performance, the application must be targeted to optimize. For more complex applications, such asSAP, forRACoptimization is more difficult. Therefore, in terms of price factors and optimization considerations, stand-alone versionOracle DB + HAsoftware is still a better solution for some scenarios. For other supportClusterthe application software of the technology can also be referred to as above analysis.

The so-called function perfect degree, refers to the application softwareClusterthe extent to which the function applies. Any technology has its limitations, although some applications have introduced the new version of theClusterfunction, but it doesHAThe support of the function is limited, and it needs to be further perfected. InTomcatas an example, althoughTomcatBack in5.0.xthe release is the introduction of the integrationSessioncluster functionality, but the main implementation is load BalancingLBaspects of the functionality, while for highly availableHAthe support and configuration are different. Tomcatof theHAconfiguration needs to be considered:Apacheis to usehttpthe direct agent orAJPAgent? whether to useSticky SessionSticky Sessions? TOMCATEnd is usedmemcached Session ManagerImplementSESSIONof shared storage and access? MSMwhether to useStickyway? and other aspects. In other words, the implementation of high availability without manual intervention (automatic detection of failures, automatic switching, automatic recovery services) is limited and difficult to configure.

based on the above two points, HA the software is still alive at the moment.

And we should also see that for a long time HA The software also has some problems in the application practice. Some problems have even seriously affected their reputation and marketing. As an example of relatively "tall" PowerHA(HACMP), the proportion configured on the IBM POWER mini-machine is quite high , basically the server for the database if the number of more than 2 units, will be purchased. The cumulative Installed capacity of these years has been very large, the technical staff of the evaluation is also very high. But what is the actual application effect? Can i automatically switch and restore services when a failure occurs? There are many examples of unsuccessful transitions. The main problem is not the problem of the product itself, but the planning, implementation and operation of the problem.

First,PowerHAcan automatically detect four types of faults: Network card failure, networking failure, node failure, application failure. However, the detection of application failures requires a customized configuration. IBMto facilitate the detection of commonly used applications, a single product is providedSmart Assist, inPowerHA v6.1it used to be purchased separately. And actually in China basically no users buy separatelySmart Assist(the user certainly does not understand, but is not only because the pre-sale did not recommend, but also has the implementation to charge the factor). Smart Assistcan helpHAthe implementers are easily configured to automatically monitorOracle,DB2, was,SAP,FilenetThe process of common applications (with different versions, the scope of support is also different, with fewer apps supported in earlier versions). NoSmart Assist, you can also manually compile the application monitoring scripts, and configure thePowerHAto implement monitoring of the application. But this requires an understanding of the application and the need to work with the people who are responsible for the application. The crux of the matter is that this is a separate charge item. Most of the agents do not have the ability to implement, and manufacturers must charge at least tens of thousands of implementation fee.

not only the implementation of the application monitoring mentioned above, in the implementation of basic functions sometimes can not guarantee the quality. Manufacturers to implement high fees, and agents of the vicious competition, the results are often free of HA implementation fee. In the case of basic users do not understand, the quality of free implementation is difficult to guarantee. Although IBM has been actively training agents, and a complete set of agency technology certification system, but the implementation of the end-user is not actually monitored.

In addition, we know that as important as the implementation of operations, and users because of lack of training and English reading problems, often can not properly maintain HA System. For example, when adding a shared storage volume LV or shared file system FS , there is no DARE through HA Dynamic configuration update mechanism and C-spoc Single Point control tool to operate. Causes shared configuration information to be inconsistent on different HA nodes. Once you need to switch, it's not working properly.

Mentioned above on these issues, believe in different HA software and different vendors will have more or less, and some are IT common problems in the process of system construction. It is gratifying that these years with the development of China's it market environment, the user's it operations team more and more mature,It staff quality is constantly improving , these issues have been or are being addressed. We believe that the high availability of our IT Infrastructure and other technologies will continue to improve and reach new heights.

This article is from the Rongxin Infrastructure Optimization blog, so be sure to keep this source http://rongxin.blog.51cto.com/8955502/1431585