PolarDB開發(fā)者大會：PolarDB在線數(shù)據(jù)實(shí)時(shí)分析加速

王劍英算力密度(CPU/Data)1.單SQL多核計(jì)算擴(kuò)展能力2.更高執(zhí)行效率的算力密度(CPU/Data)1.單SQL多核計(jì)算擴(kuò)展能力2.更高執(zhí)行效率的SQL算子單DB實(shí)例數(shù)據(jù)量大企業(yè)客戶單實(shí)例數(shù)據(jù)平均超過1TB行業(yè)頭部客戶單實(shí)例數(shù)據(jù)量在10~100TB級別緩存密度(緩存密度(Memory/Data)中小客戶傾向超配緩存，緩存數(shù)據(jù)比>10%大用戶緩存數(shù)據(jù)比<5%，更依賴IO能力0.251.更精準(zhǔn)的索引訪問路徑2.更高效率的IO訪問算法0.150.050.10.20Col-indexCol-index●共享存儲一寫多讀架構(gòu)下的行列混合存儲●基于Redo物理日志復(fù)制的實(shí)時(shí)行列數(shù)據(jù)同步●基于執(zhí)行代價(jià)的透明行列分流及節(jié)點(diǎn)間負(fù)載均衡●分析節(jié)點(diǎn)支持按需快速ServerlessCol-index 4000列式索引上的向量化算子VS行存單核計(jì)算吞吐400單表ScanScaleAggJoin行索引列索引100GTotal100GTotaltime(s)*以上結(jié)果依據(jù)客戶真實(shí)場景數(shù)據(jù)估算，最終效果以實(shí)際產(chǎn)品和場景測試數(shù)據(jù)為準(zhǔn)DispatcherHybridOptimizerSerializerHybridExecutorExchangeManagerHybridExecutorDispatcherHybridOptimizerSerializerExchangeManager列存索引彈性多機(jī)并行行列混存上的多機(jī)并行執(zhí)行時(shí)間DispatcherHybridOptimizerSerializerHybridExecutorExchangeManagerHybridExecutorDispatcherHybridOptimizerSerializerExchangeManager列存索引彈性多機(jī)并行行列混存上的多機(jī)并行執(zhí)行時(shí)間TPC-H@1TB行列混存上的彈性多機(jī)并行（ePQ）行列混存下的執(zhí)行優(yōu)化行列混合執(zhí)行GroupByGroupBy行列動態(tài)執(zhí)行ColumnBasedPlanRowColumnBasedPlanRowBasedPlanPhysicalPlanLogicalPlanCostCostBasedPlanChosenResetExecutionContextMonitorCheckExecuteRowResetExecutionContextMonitorCheckExecuteRowBasedPlanExecuteExecuteColumnBasedPlanTableScanTableScanIndexIndexScan*以上結(jié)果依據(jù)客戶真實(shí)場景數(shù)據(jù)估算，最終效果以實(shí)際產(chǎn)品和場景測試數(shù)據(jù)為準(zhǔn)OLTPOLAPOLTP/ 8060400復(fù)雜SQL查詢耗時(shí)行存列存1600140012001000800600400200011數(shù)據(jù)可見性延遲1500原方案HTAP 0歷史訂單存儲成本41原方案HTAP1行存數(shù)據(jù)12行存數(shù)據(jù)23行存數(shù)據(jù)3列式數(shù)據(jù)1列式數(shù)據(jù)1列式數(shù)據(jù)2列式數(shù)據(jù)2?復(fù)雜SQL查詢提升20倍?數(shù)據(jù)可見效延遲下降超百倍*以上結(jié)果依據(jù)客戶真實(shí)場景數(shù)據(jù)估算，最終效果以實(shí)際產(chǎn)品和場景測試數(shù)據(jù)為準(zhǔn)CoodinatorWorker關(guān)鍵特性?MPP支持交互式分析?BSP模型支持大規(guī)模吞吐型計(jì)算?統(tǒng)一內(nèi)存池?CoodinatorWorker關(guān)鍵特性?MPP支持交互式分析?BSP模型支持大規(guī)模吞吐型計(jì)算?統(tǒng)一內(nèi)存池?向量化執(zhí)行，最新指令集應(yīng)用?Adaptive動態(tài)感知自適應(yīng)優(yōu)化性能提升?離線批處理能力(SpillDisk)語法統(tǒng)一?支持MySQL標(biāo)準(zhǔn)語法?擴(kuò)展Hive語法支持?開源數(shù)據(jù)格式支持DistributedExecutionEngineWorkloadManagerDAGSchdulerTaskManagerParallelEngineExecutionEngineMemoryManagerPhysicalPlanRelationOperatorsFunctionsCode-genExecutionModelDataLayoutMemoryLayoutMemoryPool/CacheWorkload●zero-ETL零同步成本一鍵建倉●海量數(shù)據(jù)實(shí)時(shí)高并發(fā)在線分析●結(jié)構(gòu)化數(shù)據(jù)/非結(jié)化數(shù)據(jù)，湖倉一體關(guān)鍵指標(biāo)3.99X性價(jià)比TPC-性能實(shí)時(shí)大規(guī)模數(shù)倉ADBMySQL核心技術(shù)優(yōu)勢PolarDB+實(shí)時(shí)大規(guī)模數(shù)倉ADBMySQL核心技術(shù)優(yōu)勢TransactionSegmentSegmentAnalyticsSegmentSegmentMasterSecondarySegmentOSSOSS（ColumnStoreData）Primary-ROPrimary-RWPrimary-ROPrimary-RWPolarPolarStore（HybridStoreData）*