BI Forum 2009



 Principy architektury MPP datového
 skladu
                       Václav Hubka - Hewlett-Packard

26. listopadu 2009

   © 2009 Hewlett-Packard Development Company, L.P.
   The information contained herein is subject to change without notice.
Agenda

    • Spotřebiče pro enterprise datové sklady (EDWH)
    • Principy návrhu datového skladu na platfrormě
      „EDWH spotřebičů“
    • Představení Operational DWH
    • Architektura MPP datového skladu pro Operational
      DWH




    27 November
2   2009
Data Warehouse
Appliances




© 2009 Hewlett-Packard Development Company, L.P.
The information contained herein is subject to change without notice.
DWH appliances
•   Provide:
       • Systems that are packaged (tightly integrated stack, bundled, balanced, pre-
         tuned, pre-installed; single support contact) and optimized for BI workloads
       • Low maintenance and support through a single source
       • Fast installation
       • Integrated management and automated system administration. Increased
         functionality can not mean increased administrative complexity.
       • Easy incremental expansion
       • Guaranteed performance for specific purposes, use of new technologies to
         drive high performance
       • Lower TCO and faster ROI
•   Have established market acceptance
       • 10TB or more in first phase is common today
       • Have begun to host EDWs
       • Satisfy real-time, on-demand, and/or Operational BI requirements outside of
         EDW
       • 38% of TDWI research survey respondents have deployed, are currently
         evaluating, or plan to evaluate soon
Data Warehouse Appliance Pros
                           What do you think is the leading benefit of
                                a data warehouse appliance?
                                                6% = Other
                                      6% = Low cost
8% = Easy incremental expansion
                                                                  39% = Pre-tuned for
              11% = Fast installation                              data warehousing


12% = Reduced system integration
                                                             18% = Fast query performance




Sources: TDWI Tech Survey, August 2005, 119 responses
      TDWI Tech Survey, February 2007, 112 responses
Traditional data warehouse approach
               Physical            Index
     Analyze              Load               Query   Ongoing
               database             and
      data                data               data     tuning
                design           aggregate



•   Performance is dependent on getting a good
    physical design
•   Time to market with new data is limited by skills
    and resources
•   Query performance is poor when queries don’t
    take advantage of the design (no index scans)
Neoview platform performance
      Load      Query
      data      data



•   High-power-to-data ratio can operate without database
    tuning
    − Load a TB in 1 hour
    − Scan a TB in 30 seconds
•   Neoview features take performance beyond scanning
    − Next-generation optimizer can resolve complex queries efficiently
    − pMesh dual switch fabrics provide massive bandwidth between
      nodes
    − “Skewbusting” technology to resolves the traditional skew issues
      of MPP system
Simply put…
•   Overpowering a workload with inexpensive power
    has many benefits
    − The ability to perform queries that no one anticipated
    − “Load-and-go” simplicity of design
    − Reduced indices, materialized views, etc., to manage
      and tune
    − Enough power to quickly drop, reload, and restructure
      tables
What the appliance model is not good
at…
•   Random I/O
    − Scan a 1 TB table to find 1,000 rows using 256 inexpensive
      processors in 30 seconds
    − Same access with an index takes a fraction of a second using only
      one of the 256 inexpensive processors
•   Common aggregations
    − Brute force does wonders for aggregating data on the fly so that
      you don’t need to prebuild materialized views
    − The same brute force can build a MV quickly, reducing CPU
      consumption 1,000x at runtime
•   Both limit concurrency
    − How many times can one table scan in a day?
Beyond the database appliance with
the Neoview platform
                              Improve       10–1,000x faster
      Load          Query
                              database
      data          data
                               design    Much higher concurrency


     Performance repository



•   Outperforms a pure appliance with little or no design
    decisions
•   Permits you to “graduate” tables to a more enhanced
    design that supports extremely high concurrency
•   Mixed workload support allows both optimized and
    nonoptimized workloads to coexist
Operational DWH




© 2009 Hewlett-Packard Development Company, L.P.
The information contained herein is subject to change without notice.
Operational BI
Know your customer and                                                                                Timely, operational decisions
                                             Call                                     Operating
 partners                                   center                                    priorities      • Real-time analytics
• 1000s of concurrent users                                                                           • 24 x 7 availability
• Single version of the truth
                                 Self-                                                             Market
                                service                                                            signals
                                                Rich                       Responding
                                            interactions                   to business
                                             across all                     events as
                                               touch          Next             they
                                               points                         occur
                         Suppliers                          generation                                Real-Time
                                                                                                       streams
                                                              EDW

                                                     Leveraging information as a
                                                           strategic asset
Richer data inputs                    Web

• Large data volumes
• Complex, mixed workloads
                                                     Operational       Unstructured
                                                      System              data




12
Performing multiple decisions by a large
 number of users characterizes operational
 BI

      High

              Strategic             Operational
              Business              Business Intelligence
              Intelligence
Collective
impact of
decisions
              Tactical Business      Tactical Business
              Intelligence           Intelligence
      Low


             Low       Number of decisions          High


 13
The evolution of data warehouses in
operational BI environments
              Traditional BI               Operational BI

     Back room analysis          Built into business processes

     Reporting                   Automating actions

     Offline batch updates       Continuous online updates

     Departmental data marts     Enterprise-wide resource
     Few users doing strategic   Thousands of users performing many
     analysis                    types of tasks
     Availability not critical   Mission-critical




14
HP Neoview & Operational Business Intelligence –
    Real-time insight for your business
•   Enterprise data warehouse, for large enterprise operational needs
•   Scales to thousands of users, terabytes of data
•   Rapidly deployed, easily managed, compatible with existing BI tools
•   Integrated solution, HP innovation and flexible- standards-based
    components
         Data Integration               Enterprise Data Warehouse           Query Tools


                                           HP Neoview
                                              Integrated
                            Real time                          Concurrent
                            Updates           Hardware           Users

                                                  OS
                                                DBMS
                                                  Mixed
                                                 Workloads




    15
Architected for availability, scalability,
    and performance
                   •   Shared-nothing MPP
                       − Each processor a unit of parallel work
                   •   Database virtualization
 BI client




                       − Data transparently hashed across all disks
                   •   Parallel query execution
                       − Queries divided into subtasks and executed in
                         parallel with results streamed through memory
                   •   Real-time data warehousing
                       − Mixed workload & transactional heritage
ETL clients




                   •   Unrivaled availability
                       − Continuously available in spite of any single
                         point failure; online database operations
                   •   Extreme processing power
                       − 1 Intel® Itanium® processor to 2 RAID 1
                         volumes

              16
Neoview is designed
for changing customer requirements
                SLA: 20 min/2 hours                                 SLA: 5 Seconds
                                                                    Neoview: 1.6sec
                Neoview: 2min to 46min

                                                             Report
                                        Analytical           queries
                                        queries              300 concurrent
                                                                                                 Neoview: 0.5sec
                                                             1.85 million
                                        120 concurrent
                                                             queries
                                        220 queries                                               SLA: 2
                                                                                  Tactical-1
                                                                                  320 concurrent  Seconds
SLA: 10 min         Online                                                        3.6 million
                                                                                  queries
                    ingest
Neoview: 7min       7.5m rows of data        91 billion rows
                    every 10 min
                                             of data (20TB)
                                                                              Tactical-2
                                    Adhoc                                     350 concurrent
                                    surprise                                  6.4 million
                                    4 concurrent                              queries
                                    8 queries
                                                         Tactical-3                             SLA: 2 Seconds
                                                         400 concurrent
                                                         13.4 million                            Neoview: 200ms
                                                         queries

All workloads run
concurrently                                             SLA: 200 ms
                                                         Neoview: 167ms
Architecture of an MPP
DWH – HP Neoview




 © 2009 Hewlett-Packard Development Company, L.P.
 The information contained herein is subject to change without notice.
Neoview segment architecture
           Blade                                                                                                                                                                    Blade

        Node 1     Node 2     Node 3      Node 4      Node 5      Node 6      Node 7      Node 8      Node 9      Node 10     Node 11     Node 12     Node 13     Node 14     Node 15       Node 16




         P01,P02    P03,P04    P05 P06      P07,P08     P09,P10     P11,P12     P13,P14     P15,P16     P17,P18     P19,P20     P21,P22     P23,P24     P25,P26     P27,P28       P29,P30     P31,P32

         B27,B30    B05,B28     B01,B31     B02,B32     B03,B06     B04,B13     B07,B09     B08,B10     B11,B14     B12,B21     B15,B17     B16,B18     B19,B22     B20,B29       B23,B25      B24,B26




        X Fabric                                                                                                                                                                             Y Fabric




                        CS                                PS                                                                                   PS                                 CS



• 16 nodes per                                 P01                                                                                                  P14                       •   RAID1 (mirrored)
  segment                                                                                                                                                                         disk protection
• Dual active (X,Y)                            M01                                                                                                  M14                       •   Active reading from
  interconnect                                                                                                                                                                    both RAID1 copies
  fabrics                                      P15                                                                                                  P28
                                                                                                                                                                              •   Separate controller
• Multiple fat I/O
  pipes                                                                                                                                                                           writes for integrity
                                               M15                                                                                                  M28
                                                                                                                                                                              •   End-to-end disk
• Dual cluster
  switches for                                 P29                                                                                                  P42
                                                                                                                                                                                  checksum integrity
  inter-segment
  I/O                                          M29                                                                                                  M42

                                          19
Neoview multi-segment architecture
Neoview Segment                                                 Neoview Segment




FT Clustered Mesh Fabric                                            1 to 16 segments


Neoview Segment                                                Neoview Segment




                           •   Active dual fault tolerant fabrics
                           •   Multi-layered clustering
                               (>128p)
                           •   500 MB/sec dedicated links
                           •   Each segment adds bandwidth
                           •   Cross sectional bandwidth up
                   20          to 128 GB/sec
Highly Parallelized Database
                    • The key is transparently hashed to identify data
Table Table Table     placement
  A     B     C
                    • Balanced data distributions across all disks
                    • Balanced SQL execution across all processors
                    • Table, index, and materialized view support

                        Partitioning key

                                           Hash of partitioning key




21   -
                                                                         6 - 21
Neoview Shared Nothing Architecture




22
Indexed Clustering for Performance
− Data indexed for fast access by the clustering key
− Clustered data for fast sequential access

 Partitioning determined
                           Hash by     Data indexed and clustered
 by hash of partitioning
                           order #          by clustering key
            key

                                                     Cluster by
          Order                               order date, order number
                                                    Cluster by
         Line item
                                                    order date,
                                                  order number,
                                                   item number

                           Hash by
                           order #

23   -
                                                                  6 - 23
Thank you


Questions?




© 2009 Hewlett-Packard Development Company, L.P.
The information contained herein is subject to change without notice.
Co-locating Index and Base Table Data
− Eliminates cross-processor messaging overhead
− Fast and efficient indexing for query speed-up


                         Hash by
                         order #



         Index on                            Cluster by
         line item                          item number
         Line item                            Cluster by
                                              order date,
                                            order number,
                                             item number
                         Hash by
                         order #

25   -
                                                            6 - 25
Parallel UOW drives MPP performance
                                               •   Measured performance
  Itanium 2 processor   Itanium 2 processor        − Scan: 286 MB/sec/CPU
                                                     2.34 GB/sec/segment
                                                   − Ingest: 1MB/sec/CPU
                  SCAN                               to 256 CPUs using 3 loaders
                                                   − Extract: 2.5MB/sec/CPU
                                                     to 64 CPUs
    Data
   Data                         Data
                               Data
                              Data
                                                   − Insert: 1MB/sec/connection
  Data
    Management                  Management
   Management
  management                   Management
                              management
                                                     at 128 connections
                                                   − Fetch: 1.5MB/sec/connection
                                                     at 128 connections
LDV1-P   LDV1-B              LDV3-P   LDV3-B



RAID 1                       RAID 1


                        26

BI Forum 2009 - Principy architektury MPP datového skladu

  • 1.
    BI Forum 2009 Principy architektury MPP datového skladu Václav Hubka - Hewlett-Packard 26. listopadu 2009 © 2009 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
  • 2.
    Agenda • Spotřebiče pro enterprise datové sklady (EDWH) • Principy návrhu datového skladu na platfrormě „EDWH spotřebičů“ • Představení Operational DWH • Architektura MPP datového skladu pro Operational DWH 27 November 2 2009
  • 3.
    Data Warehouse Appliances © 2009Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
  • 4.
    DWH appliances • Provide: • Systems that are packaged (tightly integrated stack, bundled, balanced, pre- tuned, pre-installed; single support contact) and optimized for BI workloads • Low maintenance and support through a single source • Fast installation • Integrated management and automated system administration. Increased functionality can not mean increased administrative complexity. • Easy incremental expansion • Guaranteed performance for specific purposes, use of new technologies to drive high performance • Lower TCO and faster ROI • Have established market acceptance • 10TB or more in first phase is common today • Have begun to host EDWs • Satisfy real-time, on-demand, and/or Operational BI requirements outside of EDW • 38% of TDWI research survey respondents have deployed, are currently evaluating, or plan to evaluate soon
  • 5.
    Data Warehouse AppliancePros What do you think is the leading benefit of a data warehouse appliance? 6% = Other 6% = Low cost 8% = Easy incremental expansion 39% = Pre-tuned for 11% = Fast installation data warehousing 12% = Reduced system integration 18% = Fast query performance Sources: TDWI Tech Survey, August 2005, 119 responses TDWI Tech Survey, February 2007, 112 responses
  • 6.
    Traditional data warehouseapproach Physical Index Analyze Load Query Ongoing database and data data data tuning design aggregate • Performance is dependent on getting a good physical design • Time to market with new data is limited by skills and resources • Query performance is poor when queries don’t take advantage of the design (no index scans)
  • 7.
    Neoview platform performance Load Query data data • High-power-to-data ratio can operate without database tuning − Load a TB in 1 hour − Scan a TB in 30 seconds • Neoview features take performance beyond scanning − Next-generation optimizer can resolve complex queries efficiently − pMesh dual switch fabrics provide massive bandwidth between nodes − “Skewbusting” technology to resolves the traditional skew issues of MPP system
  • 8.
    Simply put… • Overpowering a workload with inexpensive power has many benefits − The ability to perform queries that no one anticipated − “Load-and-go” simplicity of design − Reduced indices, materialized views, etc., to manage and tune − Enough power to quickly drop, reload, and restructure tables
  • 9.
    What the appliancemodel is not good at… • Random I/O − Scan a 1 TB table to find 1,000 rows using 256 inexpensive processors in 30 seconds − Same access with an index takes a fraction of a second using only one of the 256 inexpensive processors • Common aggregations − Brute force does wonders for aggregating data on the fly so that you don’t need to prebuild materialized views − The same brute force can build a MV quickly, reducing CPU consumption 1,000x at runtime • Both limit concurrency − How many times can one table scan in a day?
  • 10.
    Beyond the databaseappliance with the Neoview platform Improve 10–1,000x faster Load Query database data data design Much higher concurrency Performance repository • Outperforms a pure appliance with little or no design decisions • Permits you to “graduate” tables to a more enhanced design that supports extremely high concurrency • Mixed workload support allows both optimized and nonoptimized workloads to coexist
  • 11.
    Operational DWH © 2009Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
  • 12.
    Operational BI Know yourcustomer and Timely, operational decisions Call Operating partners center priorities • Real-time analytics • 1000s of concurrent users • 24 x 7 availability • Single version of the truth Self- Market service signals Rich Responding interactions to business across all events as touch Next they points occur Suppliers generation Real-Time streams EDW Leveraging information as a strategic asset Richer data inputs Web • Large data volumes • Complex, mixed workloads Operational Unstructured System data 12
  • 13.
    Performing multiple decisionsby a large number of users characterizes operational BI High Strategic Operational Business Business Intelligence Intelligence Collective impact of decisions Tactical Business Tactical Business Intelligence Intelligence Low Low Number of decisions High 13
  • 14.
    The evolution ofdata warehouses in operational BI environments Traditional BI Operational BI Back room analysis Built into business processes Reporting Automating actions Offline batch updates Continuous online updates Departmental data marts Enterprise-wide resource Few users doing strategic Thousands of users performing many analysis types of tasks Availability not critical Mission-critical 14
  • 15.
    HP Neoview &Operational Business Intelligence – Real-time insight for your business • Enterprise data warehouse, for large enterprise operational needs • Scales to thousands of users, terabytes of data • Rapidly deployed, easily managed, compatible with existing BI tools • Integrated solution, HP innovation and flexible- standards-based components Data Integration Enterprise Data Warehouse Query Tools HP Neoview Integrated Real time Concurrent Updates Hardware Users OS DBMS Mixed Workloads 15
  • 16.
    Architected for availability,scalability, and performance • Shared-nothing MPP − Each processor a unit of parallel work • Database virtualization BI client − Data transparently hashed across all disks • Parallel query execution − Queries divided into subtasks and executed in parallel with results streamed through memory • Real-time data warehousing − Mixed workload & transactional heritage ETL clients • Unrivaled availability − Continuously available in spite of any single point failure; online database operations • Extreme processing power − 1 Intel® Itanium® processor to 2 RAID 1 volumes 16
  • 17.
    Neoview is designed forchanging customer requirements SLA: 20 min/2 hours SLA: 5 Seconds Neoview: 1.6sec Neoview: 2min to 46min Report Analytical queries queries 300 concurrent Neoview: 0.5sec 1.85 million 120 concurrent queries 220 queries SLA: 2 Tactical-1 320 concurrent Seconds SLA: 10 min Online 3.6 million queries ingest Neoview: 7min 7.5m rows of data 91 billion rows every 10 min of data (20TB) Tactical-2 Adhoc 350 concurrent surprise 6.4 million 4 concurrent queries 8 queries Tactical-3 SLA: 2 Seconds 400 concurrent 13.4 million Neoview: 200ms queries All workloads run concurrently SLA: 200 ms Neoview: 167ms
  • 18.
    Architecture of anMPP DWH – HP Neoview © 2009 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
  • 19.
    Neoview segment architecture Blade Blade Node 1 Node 2 Node 3 Node 4 Node 5 Node 6 Node 7 Node 8 Node 9 Node 10 Node 11 Node 12 Node 13 Node 14 Node 15 Node 16 P01,P02 P03,P04 P05 P06 P07,P08 P09,P10 P11,P12 P13,P14 P15,P16 P17,P18 P19,P20 P21,P22 P23,P24 P25,P26 P27,P28 P29,P30 P31,P32 B27,B30 B05,B28 B01,B31 B02,B32 B03,B06 B04,B13 B07,B09 B08,B10 B11,B14 B12,B21 B15,B17 B16,B18 B19,B22 B20,B29 B23,B25 B24,B26 X Fabric Y Fabric CS PS PS CS • 16 nodes per P01 P14 • RAID1 (mirrored) segment disk protection • Dual active (X,Y) M01 M14 • Active reading from interconnect both RAID1 copies fabrics P15 P28 • Separate controller • Multiple fat I/O pipes writes for integrity M15 M28 • End-to-end disk • Dual cluster switches for P29 P42 checksum integrity inter-segment I/O M29 M42 19
  • 20.
    Neoview multi-segment architecture NeoviewSegment Neoview Segment FT Clustered Mesh Fabric 1 to 16 segments Neoview Segment Neoview Segment • Active dual fault tolerant fabrics • Multi-layered clustering (>128p) • 500 MB/sec dedicated links • Each segment adds bandwidth • Cross sectional bandwidth up 20 to 128 GB/sec
  • 21.
    Highly Parallelized Database • The key is transparently hashed to identify data Table Table Table placement A B C • Balanced data distributions across all disks • Balanced SQL execution across all processors • Table, index, and materialized view support Partitioning key Hash of partitioning key 21 - 6 - 21
  • 22.
    Neoview Shared NothingArchitecture 22
  • 23.
    Indexed Clustering forPerformance − Data indexed for fast access by the clustering key − Clustered data for fast sequential access Partitioning determined Hash by Data indexed and clustered by hash of partitioning order # by clustering key key Cluster by Order order date, order number Cluster by Line item order date, order number, item number Hash by order # 23 - 6 - 23
  • 24.
    Thank you Questions? © 2009Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
  • 25.
    Co-locating Index andBase Table Data − Eliminates cross-processor messaging overhead − Fast and efficient indexing for query speed-up Hash by order # Index on Cluster by line item item number Line item Cluster by order date, order number, item number Hash by order # 25 - 6 - 25
  • 26.
    Parallel UOW drivesMPP performance • Measured performance Itanium 2 processor Itanium 2 processor − Scan: 286 MB/sec/CPU 2.34 GB/sec/segment − Ingest: 1MB/sec/CPU SCAN to 256 CPUs using 3 loaders − Extract: 2.5MB/sec/CPU to 64 CPUs Data Data Data Data Data − Insert: 1MB/sec/connection Data Management Management Management management Management management at 128 connections − Fetch: 1.5MB/sec/connection at 128 connections LDV1-P LDV1-B LDV3-P LDV3-B RAID 1 RAID 1 26