1 Performance Test summary of Telecom resource management systemChen Jianhui 2007-7-30
1.1 Technical problems and solutions
1.1.1 LoadRunner java Script unique parameter issuesWhen the static synchronized keyword was not used in the Dll,java script that originally used JNI to invoke Delphi, only 2 concurrent users were able to error the private static native synchronized String getUrl (); static{//System.loadlibrary ("Pairconn"); System.loadlibrary ("Appserverdll"); }, plus the static synchronized for complex DLLs in high concurrency or error, in the DLL shared memory mutex semaphore problem still exists, the Internet has mentioned that JNI for multi-threaded support is not very good, and finally discard the JNI interface, using the call server-side EJB to get the parameters. Originally DLL with ADO directly from the database temporary table read parameter data, the test process found that read parameters occupy a large number of database resources, and finally to the temporary table data generated data files, DLLs and EJBS directly read the data file, because the data file is relatively large, When reading parameters, you cannot load all parameter data at once, and share file pointers between multiple threads.
1.1.2 Weblogic Deployment Issues
1.1.2.1 Creating a domainProduct mode for creating domain
1.1.2.2 Update PackageCompletely update the stop instance #!/bin/sh Cd/app/bea/user_projects/domains/rms7002/bin by removing the cache of the package and then restarting the instance./stopweblogic.sh & Delete Cache script # !/bin/sh Rm-r/app/bea/user_projects/domains/rms7001/servers/adminserver Restart instance script #!/bin/sh cd/app/bea/user_projects/ domains/rms7001 rm nohup.out nohup./startweblogic.sh & View Console output #!/bin/sh tail-f/app/bea/user_projects/domains/ Rms7002/nohup.out
1.1.2.3 Java Virtual machine configuration-xms1024m–xmx1024m maximum 2G permanent domain too small to cause EJB to be unable to deploy –xx:maxpermsize=256m
1.1.2.4 Connection PoolThe initial connection and the maximum number of connections are set to the same, currently set to 80
1.1.2.5 EJB ConfigurationHigh-and-low EJB needs to be modified to modify Xdoclet's EJB annotations <pool> <max-beans-in-free-pool>1000</max-beans-in-free-pool> <initial-beans-in-free-pool>20</initial-beans-in-free-pool> </pool>
1.1.3 Database
1.1.3.1 Large Data scale partitionAccording to 5 million records of a district, Pair_conn 13 partitions, connector table 34 partitions, respectively partitioned
1.1.3.2 Large concurrency operation tablen Delete the useless index and all foreign key n increase the Initrans parameter of the table and index, the connector table is changed to 24, the other high and publish the Freelist parameter to increase the table and index by N. freelists All indexes and primary keys are put into index table space n All tables and indexes are changed to nologging
1.1.3.3 Data Import
1.1.3.3.1 Rebuilding a user's scriptDrop user RMS cascade; CREATE USER ' RMS ' profile ' default ' identified by ' XXXXX ' default tablespace ' trmsdata ' account UNLOCK; GRANT DELETE any TABLE to "RMS"; GRANT EXECUTE any PROCEDURE to "RMS"; GRANT INSERT any TABLE to "RMS"; GRANT Select any SEQUENCE to "RMS"; GRANT Select any TABLE to "RMS"; GRANT UNLIMITED tablespace to "RMS"; GRANT UPDATE any TABLE to "RMS"; GRANT "CONNECT" to "RMS"; GRANT "DBA" to "RMS";
1.1.3.3.2 Import Data ScriptImp rms/xxxxx@trmstest file=rmsdb.dmp full=y commit=y ignore=y buffer=1024000 feedback=100000
1.1.3.3.3 Analysis TableOracle Table Analysis improves performance after importing data large table with separate session increase sort_area_size increase analysis speed alter session set sort_area_size = 100000000; Alter session Set sort_area_retained_size = 100000000; Alter session Set Sort_multiblock_read_count = 128; Alter session Set Db_file_multiblock_read_count = 128; Analyze table RMS. Atom_res_serv_ins compute statistics;
1.1.4 Procedural Issues
1.1.4.1 EJB Comment is not configured correctlyMost helper-bean-id are inconsistent with the bean-name in the spring configuration file
1.1.4.2 Hibernate, JDBC, storage processL batch operation changed to use the storage process faster than with JDBC nearly 100 times times L hibernate efficiency is the lowest, the device increased originally used hibernate when the WebLogic server CPU is high L database connection leaks, not as required Jdbctemplete, instead of writing the JDBC code yourself, there is a large number of such problems in the program storage procedure Call
1.1.4.3 SQL OptimizationL is null does not use the index will only do a full table scan L where conditions in the result set small conditions are placed behind L for large table subqueries more efficient than table connection L complex query statement with PL/SQL view execution plan, see Yes Whether there is a valid index, there is a full table scan for large tables
1.1.5 Operating System
1.1.5.1 Adjusting the oracle9i database host kctune-h nproc=4096 kctune-h strmsgsz=65535 kctune-h dnlc_hash_locks=512 kctune-h ksi_alloc_ max=32768 kctune-h max_thread_proc=256 kctune-h maxdsiz=1073741824 kctune-h maxdsiz_64bit= 2147483648 kctune-h maxssiz=134217728 kctune-h maxssiz_64bit=1073741824 kctune-h maxswapchunks= 16384 kctune-h maxtsiz=0x4000000 kctune-h maxtsiz_64bit=0x40000000 kctune-h maxuprc=3000 kctune-h& nbsp;maxusers=4096 kctune-h msgmap=4098 kctune-h msgmni=4096 kctune-h msgseg=16384 kctune-h msgtql=4096 kctune-h ncallout=8000 kctune-h ncsize=34816 kctune-h nfile=63488 kctune-h nflocks =4096 kctune-h ninode=34816 kctune-h nkthread=7184 kctune-h nstrpty=60 kctune-h semmni=8192 kctune-h semmns=16384 kctune-h semmnu=4092 kctune-h semvmx=32768 kctune-h shmmax=16743656000 kctune-h shmmni=512 kctune-h shmseg=32 kctune-h vps_ceiling=64
1.1.5.2 weblogic9.2 Host kernel parametersKctune-h max_thread_proc=4096 kctune-h maxusers=2048 kctune-h maxfiles=2048 kctune-h maxuprc=1024 kctune-h maxdsiz=0x 40000000 kctune-h maxdsiz_64bit=0x80000000
1.1.5.3 OS PatchesOS patch not installed, Hpjconfig discovered by HP engineers during formal testing the system does not have a Java 1.5 patch installed
