《【T112017-数据工程和技术分会场】物联网和人工智能领域内置芯片分析的意外之旅.pdf》由会员分享，可在线阅读，更多相关《【T112017-数据工程和技术分会场】物联网和人工智能领域内置芯片分析的意外之旅.pdf（53页珍藏版）》请在taowenge.com淘文阁网|工程机械CAD图纸|机械工程制图|CAD装配图下载|SolidWorks_CaTia_CAD_UG_PROE_设计图分享下载上搜索。
1、Future-proofing BI:an unexpected journey to leverageIn-Chip analytics in IoT and AIAni ManianHead of Product Strategy|TalkingdataSIMPLIFYING Business Analytics for COMPLEX Data“The key strength of Sisense is the platforms capability to easily handle and manage large and diverse datasets,and analyze
2、them in dashboards based on its proprietary In-Chip technology.”-Gartner Magic Quadrant|TalkingdataHOW IT ALL STARTED|TalkingdataWHAT DO FIVE DATA GEEK STUDENTS DREAM ABOUT?|TalkingdataWELL,BELIEVING THEYRE BADASS THEYRE DREAMING OF|TalkingdataBEER&CHIPS|TalkingdataBeerDataIN ORDER TO UNDERSTAND IN-
3、CHIPANALYTICSLETS ASSUME THAT:|TalkingdataMEMORY HIERARCHY IN MODERN CPUSL3 CacheCapacity:6MB-20MBLatency:35 CyclesL2 CacheCapacity:256KB-1MBLatency:10 CyclesL1 CacheCapacity:64KB-128KBLatency:3 CyclesCPUMain MemoryCapacity:GBs-TBsLatency:150-450 CyclesRAMRAMRAMDiskCapacity:UnlimitedLatency:1M Cycle
4、sDISK|TalkingdataSO,WHY SHOULD WE EVEN CARE?Slowdown when fetching new data to the CPUx50SlowdownMain MemoryUp tox100 x10SlowdownL3 Cachex3SlowdownL2 Cache|TalkingdataMEMORY BANDWIDTH L1 cacheHome fridge DistanceImmediateCustomerx1L2/l3 cacheShopDistanceBicycleCustomerx10RamSupermarketDistanceCar Cu
5、stomerx50DiskBreweryDistanceAirplaneCustomerIf data equals beer then data storage units equal all the places beer is kept!|TalkingdataTHERE SHOULD HAVE BEEN A SLIDE HERE.(its the beers fault)How does Sisense overcome the memory bottleneck?|TalkingdataStore all data on the DiskOnly Use RAM When a Que
6、ry RunsLoad Only the Relevant Columns in RAMHOW DOES SISENSE OVERCOME THE MEMORY BOTTLENECK?VECTORIZATIONJIT LLVM&SIMD|TalkingdataVECTORIZATION&CACHE AWARENESSL1 CacheFirst into RAMOP1004K(Values)1004K(Values)1004K(Values)Result VectorPush Back To RAM1004K(Values)SIMD REGISTERApply Operation On 4/8
7、Data Elements SimultaneouslyOPOPColumn 41004K(Values)ResultVector1004K(Values)Column 11004K(Values)Column 21004K(Values)Column 31004K(Values)|TalkingdataJIT LLVM COMPILATION WITH SIMD SUPPORTint f int a,int b)elsem0m1m2m3m0m1m2m3Return a;a0a1a2a3returna=0;a0a1a2a30000withmaskm0m1m2m3if a 0)m0m1m2m3a
8、0a1a2a310=0/=1OR&Mask Vectorf2 Vector“Customer 1”“Customer 2”OR&Mask Vectorf3 Vector“1”/”2”/”3”OR&Mask VectorL1 CacheField Vector=ValueMask Vector=True/FalseSELECT(f1=“beer1”OR f1=“beer2”)ANDFROM T1(f2=“customer1”OR f2=“customer2”)ANDWHERE(f3=“1”OR f3=“2”OR f3=“3”)AND(f4”10”OR f4=“0”OR f4=“1”)|Talki
9、ngdataNEXT:PERFORMANCE TUNING FOR MANY USERSADD INSTANCESADD HARDWAREOPTIMIZE DATA MODELHOW CAN YOU DELAY USING THESE OPTIONS?|TalkingdataPROBLEM:THE WAITING LINE TO QUERY DATAThe queue means a user wait is extended by each user in front of themUSERSSECONDSCPU|TalkingdataQUERYS BUILDING BLOCKS:THE I
10、NSTRUCTION SETS|TalkingdataCROWD SPEED:MACHINE LEARNING ARCHITECTUREBreak each query into partsStore each query part and learn Build new queries with matching parts to boost performanceQUERYEXECUTIONSPEED|TalkingdataRE-USE REPEATING INSTRUCTION SETS ACROSS QUERIES#1HOW MANY UNITS DID WE SELL?New Que
11、ry#2WHAT WERE THE MONTHLY SALES?Already calculated units sold#3WHO WERE THE TOP SALES REPS EACH MONTH?Already calculated units sold&Monthly breakdown of unitsSimilar but non-identical queries|TalkingdataMACHINE LEARNING BIWith Machine Learning BI,analytics get faster even when queries are not identi
12、cal.The more questions you throw at it-the more efficient it gets!More users=more queries=faster resultsNo matchMatch found|TalkingdataIN-CHIP=POWER+MACHINE LEARNINGLeverage the unique in-chip cache memory to perform faster than in-memoryWithout the limitation of having to load the entire model into
13、 RAMIn-Chip recognizes the CPU specs and applies its unique code to organize the query data in the CPUWhen needed again,that piece of data exists in the CPU cache,which is much faster than RAMIn-Chip machine-learns to fetch the associated compressed result sets in advanceSub-query results pre-loaded
14、 into L1 cache as compressed dataDecompressed images of that same data can be moved to the larger,but slower,L2 and L3 caches Decompression operations(read from and write to cache)are extremely fast|TalkingdataIN-CHIP TECHNOLOGYThe best engine beer can buyIn memory columnar execution modeCACHE aware
15、 query kernelCACHE awaredecompressionInstruction recycling&learning algorithmsLLVM based compiler with SIMD supportFull 64BIT supportColumnar storage|TalkingdataDataset:120M rows 28GB8 Analytical queries X 50 cyclesAggregationsGroupingTop RankingLarge intermediate results BENCHMARK SETTINGS1 601 401
16、 201 00806040200Test 1:No concurrencyTest 2:Concurrency=2Test 3:Concurrency=MaxTest 1:No concurrencyTest 2:Concurrency=2Test 3:Concurrency=MaxIN-CHIP BI BENCHMARK300%fasterSkylakeHaswellEMPOWERING GROWTH,ANYWHERE,EVERYWHERE,ON AFFORDABLE HW|TalkingdataSPEED!STRATA AWARDAnalyzing 10TB of data in 10 s
17、econds On a single node on a standard Dell Server|TalkingdataREVOLUTION:SCALE-OUT VS IN-CHIPArchitectureUsers Use CasesInterfaceTime to ImplementAvailable ResourcesOutcomeIn-ChipBusiness UsersAd-Hoc AnalyticsInteractive Dashboards,SQLShortSmallAgile Big Data AnalyticsScale-Out Data Scientists,IT,Dev
18、elopersETL,Batch Reports,Machine LearningJAVA,R,C,SQLLongBigBig Data Infrastructure|TalkingdataSOWHAT IS IT GOOD FOR?|TalkingdataFROM COMPLEXITYTO SIMPLICITY|TalkingdataTECHNOLOGY HAS NO MEANING IF IT HAS NO IMPACT ON HUMAN LIFE“If a tree falls in a forest and no one is around to hear it,does it mak
19、e a sound?”|TalkingdataOVERHYPE OF BUZZWORDS|TalkingdataTHE PERSONAL,INTELLIGENT AND CONTEXTUAL WEB|TalkingdataTHE INTERNET OF ME|TalkingdataTRANSFORMATION OF BIG DATA ANALYTICS FOR IOM|TalkingdataWE ARE ALL UNIQUE|TalkingdataLET THERE BE LIGHTS|TalkingdataTHE NEXT REALM OF BUSINESS ANALYTICS Analyz
20、ing data no longer requires being anchored to a screen Sisense Everywhere devices broadcast business KPIs to all the senses Making consumption of insights immediate and simple.|TalkingdataHOW IT ALL STARTEDThe relationship between business professionals and their KPIs(We asked hundreds of business p
21、rofessionals how they interact with their data and KPIs)|TalkingdataALMOST HALF OF ALL RESPONDENTS CHECK KPIS DAILYHow often do you check the status of your KPIs?|Talkingdata83%OF RESPONDENTS USE OR WANT TO USE COLOR CODINGDo you use color-coding in the way you display data?|TalkingdataVISUAL ALERTS
22、 ARE THE MOST EFFECTIVE,ACCORDING TO MORE THAN HALF OF RESPONDENTSWhich type of alert is best in driving you to action?|TalkingdataACCORDING TO RESPONDENTS THE FUTURE OF BI CONSUMPTION IS EVERYWHEREHow would you like to consume data in the future?|TalkingdataBI EVERYWHERERevolutionizing The Way Busi
23、ness Users Consume Data|TalkingdataIMAGINE A SISENSE WORLDImagine youre driving to work and can ask your voice-operated BI assistant:What is my sales target for today?|TalkingdataIMAGINE A SISENSE WORLDImagine being able to focus your entire team on improving customer satisfaction just by having the
24、m glance at the Sisense IoT bulb,green means on target and red means take action.|TalkingdataIMAGINE A SISENSE WORLDImagine stepping into a conference room for a quarterly business review and experiencing your data insights hovering around you.|TalkingdataSISENSE BRINGS IMAGINATION TO LIFESisense-En
25、abled is a new line of devices that present data unlike any dashboard environmentSISENSE LAMPSISENSE ENABLED ECHO|TalkingdataREDEFINING HOW WE INTERACT WITH DATA“When I see that bulb change,I get a real sense of satisfaction.Its provided a direct way for us to see how data is changing.The bulb gives
26、 me peace of mind because I can see a light change rather than monitoring a screen.”LIVE CLOSER TO YOUR DATA“Bulb is the KPI that you dont need to load up on one of your screen,its not just another browser window.Its this physical piece thats simply part of your life.Its a simple product with a powe
27、rful way of telling you whether things are going well.”RESPOND TO CHANGES IN REAL-TIME“I think I find it easier to relate to color and sound than a dashboard.I have seen a change in my behavior using these tools,specifically around time to react-understand when something is changing and going to loo
28、k at metrics to find out why.”|TalkingdataSIMPLIFYING COMPLEX DATA CONSUMPTIONMAINTAIN FOCUSKeep teams focused on a common goal and in touch with your business.GAIN CONSTANT VISIBILITYKnow whats happening,wherever you are,in an instant.STAY CONNECTEDKeep your finger on the pulse and act on whats important.|TalkingdataTHE FUTURE OF ENHANCED HUMANISMTHANKS