OracleInMemory目錄CONTENTSIn目錄CONTENTSInMemoryInMemoryInMemoryInMemory太多InDBResultMemoptimizeforMemoptimizeforTimestenIn-MemoryOracleDatabaseInMemoryBufferCache

NewIn-Memory

InMemory

在行存儲中處理DML目錄CONTENTSIn目錄CONTENTSInMemoryInMemoryInMemoryInMemoryInMemory InMemory聚Vector

HASHTable Tableallvaluesan1

InMemoryExample:ComputetotalsalespriceNet=Price+Price*TaxIn-MemoryColumnPrice+Price+Price*

In-MemoryColumnSales_Q1

InMemory

ColumnStore

securefilelobIndex

Fast

FASTSTART

InMemory：ActiveDataGuard1In-

1In-

InMemory與BUFFERCACHE訪問相比性能提高7到128與磁盤命中相比性能提高62到3259混合場景提速5到9倍（通過刪除多余索引實現(xiàn)全面基于目錄CONTENTSIn目錄CONTENTSInMemoryInMemoryInMemoryInMemory

DataIn-MemoryColumnIn-MemoryColumn

StagingStaging

3rdNormal

PerformanceSTARPre-Cal

星型模式和預(yù)先計算的InMemoryOLAP業(yè)務(wù)或數(shù)據(jù)庫評估InMemory純OLTPInMemoryIn-MemorySGA_TARGET=源SGA_TARGET+INMEMORY_SIZE（單實例SGA_TARGET=(源SGA_TARGET+INMEMORY_SIZE)*1.1PGA_TARGET=PARALLEL_MAX_SERVERS*如果無法避免寫臨時段：_smm_auto_max_io_sizeHashOutoflineCompressionAdvisor&In-Memory內(nèi)存InMemory對象InMemory

SQL>SELECT*FROMV$INMEMORY_AREAWHERECON_ID=1MB64KBSQL>selectsegment_name,inmemory_size,bytes,populate_statusfrom BYTESBYTES_NOT_POPULATED 0內(nèi)存InMemory對象InMemory執(zhí)行計劃INMEMORY

SQL>select/*+full(t)*/count(*)fromt_inmemoryt;Planhashvalue:|Id| | |Rows|Cost(%CPU)| 0|SELECT 1| (6)|00:00:01 1|SORT 1 2 TABLEACCESSINMEMORYFULL|T_INMEMORY|5307K| (6)|00:00:01內(nèi)存InMemory對象InMemory執(zhí)行計劃INMEMORY

62recursive0dbblock18consistent5physical0redo561bytessentviaSQL*Netto410bytesreceivedviaSQL*Netfrom2SQL*Netroundtripsto/from0sortssortsrows內(nèi)存InMemory對象InMemory執(zhí)行計劃INMEMORYIMscan

SQL>selectname,value,t.con_idfromv$statnamen,v$sesstatwheren.statistic#=andn.con_id=andname='IMscanandsidin(selectsidfromv$mystat); IMscan InMemory與In-Memory是sharenothing架In-Memory的表會打散到RAC數(shù)據(jù)打散方式包括ROWIDRANGE、In-Memory的DuplicateExadataInMemory與RAC環(huán)境如何高效使用InPARALLEL_FORCE_LOCAL設(shè)置為PARALLEL_DEGREE_POLICY設(shè)置為需要確保N/nodes合理規(guī)劃表分區(qū)并按照PARTITION方式分布In規(guī)劃業(yè)務(wù)訪問SERVICE對應(yīng)的節(jié)點，按照SERVICE分布InDataGuard&

In-MemoryIn-Memory對OLTP對于網(wǎng)絡(luò)交互，登錄，解析，PL/SQL，復(fù)雜函數(shù)，數(shù)據(jù)處理，ETL需要關(guān)注由于空間，連接方式，RAC架構(gòu)以及BUG導(dǎo)致的In-MemoryIn-MemoryRAC目錄CONTENTSIn目錄CONTENTSInMemoryInMemoryInMemoryInMemoryIn-MemoryColumnHot CoolAuto

In-Memoryinmemory_automatic_level=INMEMORYHotTables

Hot

Cold

12c(18c/19c):Memory訪問

WHEREPrice>1000SALESLEDLED

RowStore(BufferSALES20c:從BufferCache

ScanandfilterbyPriceusingcolumnstore

LEDIn-Memory連接處理速度提高5-10

JoinSalesandVectorVector

VectorIn-MemoryExternal

Object

向量處理JSONIn-Memory

分區(qū)（internalpartitions）與外部分區(qū)（external

CREATETABLEorders(order_idorder_dateDATE,…EXTERNALPARTITION(TYPEoracle_loaderDEFAULTDIRECTORYdata_dirACCESSPARAMETERS(..)REJECTLIMITunlimited)PARTITIONBY(partitionq1_2015valueslessthan(‘2015-04-EXTERNALLOCATIONpartitionq2_2015valueslessthan(‘2015-07-01’),partitionq3_2015valueslessthan(‘2015-10-01’),partitionq4_2015valueslessthan(‘2016-01-In-MemoryIn-MemoryIn-Mem