Tuxedo 设计优化[精选].pptVIP

BEA系统有限公司 三层架构的物理表示 三层架构的物理表示 三层架构的物理表示 Tuxedo Environment Tuxedo Traffic Flow Queueing Network Model Server Node View 分布计算的实质 – 分布资源的利用 利用分布的系统资源提高对象/服务的并发度 利用分布的应用结构提高组件的重用程度 利用分布的应用结构提高软件的容错性 利用分布的应用结构提高系统部署的灵活性 减少分布造成的Delay 减少分布造成的连接次数 减少远程调用造成的数据传输 提高分布的并发度 优化业务流程 通过应用设计实现： 用无状态的对象或服务完成流程的状态 异步调用（扇出式并发） 事件机制 通过应用服务器实现 多机集群效率, 对象和状态的复制 异步调用和事件API 负载均衡的手段 应用控制的手段 Timeout Request Routing Data compress Transaction N层分布应用体系 分布计算的实质 – Queue Network 分布计算的性能 Some Expensive Operations tpinit PK Security File Transfer Compression Domains Distributed transactions Networking 性能优化 Bottom-Up Strategy Tune OS Tune Tuxedo UBB parameters Tune DB Logic Testing/Monitoring Performance Back to step 1/2/3/4 Tuning is endless process, so set a goal for tuning! Tuning OS Make sure OS IPC parameters are configured optimally Symptoms of bad configuration IPC queues of Tuxedo processes get full Clients get exceptions Monitor with ipcs -qa (works on both Unix NT) e.g. OS Parameters shminfo_shmmni=1024 seminfo_semmns=10240 seminfo_semmsl=100 seminfo_semmni=1024 seminfo_msgmnb=65535 seminfo_msgmax=16384 seminfo_msgmni=512 Tuning Tuxedo：后台服务进程 多少进程才能有效利用资源 ？ “Too many” as well as “Too little” can negatively impact performance Too many -- more system resources such as memory, context switching, etc Too little -- IPC queues can back up, etc Tuning Methodology Start with a fewer number of processes and tune from there. Monitor the respective IPC queues and consider increasing the respective processes if queues back up Reduce the number of server processes if some of them rarely get used /Q Optimizations Use Raw disk for Queuespaces and TLOGs Use Memory queues if possible Use TMQFORWARD –f option if appropriate Transaction Optimization Each 2-Phase commit must write a record to the TLOG TLOG should be on a Raw, mirrored device Start a lot of TMSs – they will issue their I/O requests in parallel, and the OS will keep the disk working efficiently Avoid if possible Two-Phase Commit Datafl