Making Sense of Hadoop - It's Fit with Data Warehousing Solutions

Loading...

Flash Player 9 (or above) is needed to view presentations.
We have detected that you do not have it on your computer. To install it, go here.

0 comments

Post a comment

    Post a comment
    Embed Video
    Edit your comment Cancel

    Notes on slide 1

    So, what IT is attempting to do today, is to sidestep the traditional EDW flow, and side-step traditional analytics technologies in two ways.First IT is trying to shorten the path from applications to analysis – by directly feeding application data into an analytic database designed to support the frontlineAnd second, application groups are figuring out ways to analyze and transform data without moving it from the database. This is behind the recent surge of interest in data transformation tools like hadoop and map/reduce.

    Tasso

    KEY POINTS:So, there are a lot of options that you can choose from, but there are a few things that make opsw different and truly enable customers to get maximum value out of automation. With Storage, there are three 3 areas where you benefit from our entry into Storage automation. The three things are: We’re the only integrated suite. We integrate across servers, networks and storage. You can buy point tools for every piece, but then you have the integration nightmare of tying them together. If one needs to be upgraded, or swapped out with another point tool, you have to go through all the integration again. If you think about what your customers care about (the IT customer) – they care about the application – does the app work? Is the service working or not? Our integrated suite enables visualization of dependency analysis, integrated & coordinated change and troubleshooting across entire app infrastructure as opposed to just any one of these silos. This really critical – if you look at our customers – and the stats in terms of what they were able to achieve with the individual products - where they really get the exponential value and transform how they manage IT – its from the integrated suite – which enables them to manage at a whole new way: a very service oriented / customer oriented way.Two: Mapping storage to Applications. So, really helping you understand the end to end infrastructurte of each of your services, and provide a single model of the those services. The unprecedented visibility that ASAS gives administrators will save countless hours in downtime and errors. You will be able to visually inspect your server / SAN topology, visually inspect the logical data flow to your storage infrastructure.  And, where virtualization layers come into play, we will be able to show you how they correspond to physical resources and tie that together the dependent application.  You will be able to see an application from the storage point of view, where the various storage resources are coming from.   This will allow you to perform risk analysis.     Three: Opsware will give you the information you need of knowing what’s out there in your infrastructure, how it is inter-connected and how it is changing. You will have a central view of how your business services are compliant across the entire IT infrastructure – not just one silo. You will have unprecedented visibility to proactively identify issues before they become outages & minimize risk.

    Pain: Sequential insert-based data import from HDFS cluster was slow to migrate into Aster (and vice-versa).Previous solution: LinkedIn’s data transfer GUI connected to Aster via JDBC and did batch inserts. It was a sequential loading process with high overhead. LinkedIn estimated that a serialized JDBC load process would take 15 hours to transfer from HDFS to Aster. LinkedIn gave up after 2 hours, as the latency was unbearable and unacceptable to meet their SLAs. Note to Shawn: consider rations (eg 7x faster etc)Note: Hadoop over HDFS cannot pull/push in a parallel way – only Aster can control thisAster SolutionLinkedIn created a SQL/MR function called HDFS2Aster (and Aster2HDFS) that can be issued from Aster nCluster to read data from HDFS in a parallelized mannerHDFS2Aster code took <100 lines of code. Key benefits include: 1) parallelized data copy from HDFS and back to HDFS, 2) seamless/simplicity/automatated – get data from a physically disparate system (filesystem) – can just use the same SQL workflow, Resulting ImpactResource Costs (simplicity/seamlessness saves 10x time (confirm with LinkedIn) of complex manual effort every day in doing HDFS loading – development effort/time is savedPerformance (parallelization means 90 minute load times vs 900 minutes (15 hours), a 1000% performance improvementLong-term benefit – sandbox environment to play with in Aster. Currently HDFS data put into Aster for prototyping. Ability to have database users and engineering/hadoop users (flexibility). ETL in Aster, consolidation. People who know databases understand both hadoop and aster and no need to create a mapreduce job – just a simple SQL workflow – for Analytics usage in nCluster. Hammer home the *seamless* nature of SQL/MR. Basically HDFS2Aster is parallel external tables with Hadoop (simplicity, seamlessness) – alternative is multi-step and multi-tool approach with multiple failure domains. If SQL/MR fails, it’s just 1 transaction failing and you can roll back – database manages everything

    SEW:Aster Data = 2 words

    Analysts want to use SQL

    We all know that networks have become the electricity that businesses run on. It used to be that you had a simple, straight-forward network, and it took 1 to 2 people to manage it. But as business dependencies on the network have grown, the network has become much more complex. Today, as you know, networks are distributed across the country – or in your case – across the world. You have more devices, with more complex configurations, and one of the single biggest problems that large companies face is just simply having visibility into the environment. Just basic knowledge like what are all the devices I have in my network? and more detailed things like are there any critical applications dependent on this switch port? That lack of visibility and the inability to get accurate, up to date information is creating huge problems for network teams as they try to manage increasingly complex network infrastructure. Next comes compliance. The rise of compliance regulations and the mandates that you must follow means you could have one of you network engineers documenting and validating your network standarrd s for 4 weeks a quarter or you could automate that routine, extremely boring task and turn that network engineer over to roll out the new MPLS network that is going to save your IT department 600K a year. This leads me to my third point which is that we are seeing our customers under tremendous pressure to roll out new services like VoIP, wireless networks and MPLS. And, of course, this not only increases the number of devices that you have, but also increases the complexity. It used to be that 1 or 2 people could track all of this in their heads or on a simple spreadsheet, but the network has grown way too large and complex. I’ll give you an example here, I was talking with one of the largest financial organizations in the world last week, and the network manager told me that they are adding between 6 and 20 new devices a week. That’s how fast their network is growing. Build out slide that tells a story. Starts with a small network and ad-hoc processes and builds to a global network with advanced services, processes, etc.Lack of VisibilityIncomplete device inventoryUnknown versions of software MisconfigurationsNon-ComplianceAmbiguous regulations, failed audits, security vulnerabilitiesIncreasing Complexity and ScaleNetwork growth, new service roll out, flat staffing, increasing OpExNegative Business ImpactCannot see relationship between infrastructure and business servicesDeliver services to other teams automaticallyT-mob – automating validationMoves, changes, addNobody has time for processMake network higher qualityDeliver services quicklyAll across a growing, globally distributed networkComplexityIncreasingly complex infrastructure(M&A, web services, SOA adoption etc. )Couple these challenges with today’s current climate – things are accelerating and these problems are going to get worse. Blurring lines between server and network silos IT is changing from a cost center to a strategic asset85% of total budget is spent on maintenance & operations Enterprises are recognizing that IT is not just a cost center and instead can be used for company innovation & competitive leverageAccording to Forrester Research, 85% of IT budget is dedicated to operations & IT maintenance costsNetwork is becoming the platform for applications – increased dependenceFinger pointing between server & network teamsAcquisition, Congo line of devices, Compliance – Increasing network compliance adherence vigilance Notion of networking tam being able to deliver network funtions to other organizations.Think about server in staging that must be moved to production and the process involved

    13 Favorites

    Making Sense of Hadoop - It's Fit with Data Warehousing Solutions - Presentation Transcript

    1. Making Sense Of Hadoop – Its Fit with Data Warehousing Solutions
      Shawn Kung, Sr. Director Product Management, Aster Data
      Colin White, President & Founder, BI Research
      October 1, 2009
    2. Topics
      Aster Introduction
      Analyst Perspective: Colin White on Hadoop
      Introducing Aster-Hadoop Data Connector
      Case Studies: LinkedIn, ShareThis, comScore
      Summary
      2
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
    3. Aster DataCreating the Next-Generation Data Management System
      Founded in 2005 to revolutionize data analytics and management of large data volumes
      Founding team innovated on the ‘big data’ problem at Stanford University – and were joined by big data experts from Google, Oracle, and MSFT
      Aster’s first commercial product, nCluster, for big data volumes has been in market since 2007
      Since 2008 innovated on Google’s well-known MapReduce framework to transform data processing. Created patent-pending In-Database MapReduce (SQL/MR) interface
      $27M of funding from leading venture firms
      3
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
    4. Aster TodayRapid adoption of nCluster and MapReduce
      Today – the fastest growing company in the ‘big data’ market
      Powering the largest data warehouse in the Cloud (customer ShareThis)
      Powering the most data-intensive applications at MySpace, Full Tilt Poker, Akamai
      Major customers across Internet, retail, financial services
      MySpace, Coremetrics, LinkedIn, Akamai, ShareThis, comScore, Full Tilt Poker…
      Broad product line – 3 offerings
      Software-Only, Aster nCluster Appliance, and Aster nCluster Cloud Edition
      4
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
    5. Our Customers’ Challenge:Massive Increase in Data Volumes and Data Formats
      5
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
    6. New Generation Apps Compound the Problem
      6
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
      • Large data volumes
      • Rich data formats
      • Organic data growth (GB  TB  PB)
      • Data ‘sampling’ not an option
      • Granular data analysis
    7. Typical Customer Data Pipeline Challenges
      7
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
      Apps & Users
      ReportingApps
      Real-TimeData Source
      RDBMS/EDW Data Flow
      CRMApps
      Enterprise
      Data
      Warehouse
      Source
      OLTP
      Database
      Operational
      Data Store/ MDM
      Analytic Data
      Mart
      Load
      Transform &Batch Load
      Sample
      Extract
      WebApps
      Real-TimeData Source

      Mobile Apps
      Real-TimeData Source
      Limits Access to Data TB’s of data filtered down to GB’s (small samples)
      High Latency –Days to weeks to access and analyze data
      Social Networks
      Severely Restricts Analysis – Significant data movement since analytics & data store are separate + SQL restricts queries
      Structured Data Only Not designed for non-relational data
    8. Aster nCluster: Solving The Big Data Challenge
      8
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
      Reports, Analytics, Applications
      (SQL & extensions / ODBC / JDBC)
      Queries/Answers
      Data sources
      Queen Server Group
      Data
      Queries
      Worker Server Group
      Loader/Exporter Server Group
      Aster nCluster Database
      “Always Parallel” Scalable Performance
      “Always On” Resiliency
      In-Database MapReduce
      Commodity Hardware Incremental Scaling
      Structured & unstructured data
    9. Topics
      Aster Introduction
      Analyst Perspective: Colin White on Hadoop
      Introducing Aster-Hadoop Data Connector
      Case Studies: LinkedIn, ShareThis, comScore
      Summary
      9
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
    10. Making Sense of Hadoop: Its Fit with Data Warehousing Solutions
      Colin WhiteFounder and PresidentBI ResearchOctober2009
    11. Agenda
      The Big Data Problem
      Capturing, Managing and Analyzing Big Data
      The Role of Hadoop and MapReduce in Big Data Problems
      Fitting Hadoop into a Corporate Data Warehousing Environment
      Copyright © BI Research 2009
    12. Big Data
      The New York Stock Exchange generates about one terabyte of new trade data per day
      Facebook hosts approximately 10 billion photos, taking up one petabyte of storage
      Ancestry.com, the genealogy site, stores around 2.5 petabytes of data
      The Internet Archive stores around 2 petabytes of data, and is growing at a rate of 20 terabytes per month
      The Large Hadron Collider near Geneva, Switzerland, will produce about 15 petabytes of data per year
      plosone.org
      Source: “Hadoop: The Definitive Guide,” Tom White, O’Reilly Media, June 2009
      Copyright © BI Research 2009
    13. Using Data for Business Benefit: The Role of Analytics
      Top ten CIO directions, IBM global survey of 2,500 CIOs, September 2009
      Copyright © BI Research 2009
    14. Processing Big Data: The Problem
      Operationalsystems
      Analyticalsystems
      Run thebusiness
      Analyze thebusiness
      capture
      Capture?
      Manage?
      Analyze?
      decisions& actions
      actionableintelligence
      Improve thebusiness
      Collaborativesystems
      Copyright © BI Research 2009
    15. Big Data Directions: Capture and Management
      capture dataraw events,& rich content
      Low-cost commodity hardware
      Hardware/software appliances
      Virtualization
      Cloud computing
      Optimized file systems(XML, columnar, HDFS)
      Data filtering & aggregation
      Operationalsystems
      Analyticalsystems
      filtered information
      operational data & content
      data warehouse information & analytics
      Copyright © BI Research 2009
    16. Big Data Directions: Analysis
      in-flightanalysis
      Parallel computing
      Analytical RDBMSs
      In-database processing
      New analysis techniques(predictive analytics, content analytics, MapReduce)
      In-flight analysis
      Operationalsystems
      Analyticalsystems
      operational data & content
      data warehouse information & analytics
      Copyright © BI Research 2009
    17. Application Development: What is Hadoop?
      Hadoop is an Apache open source project that includes:
      MapReduce: A software framework for distributing the processing of large data files across multiple machine clusters
      HDFS: A distributed file system that stores and replicates large files across multiple machine nodes without the need for RAID storage
      HBase: A column-oriented, distributed database that runs on top of HDFS (modeled after Google&apos;s BigTable storage system)
      Hive: Tools for batch-style query, analysis and summarization of large Hadoop data files using QL queries
      Pig: A platform that provides at alternative to MapReduce for analyzing large Hadoop data files using high-level data analysis language programs
      Other subprojects including Avro, ZooKeeper, Chukwa and Hadoop Common
      Copyright © BI Research 2009
    18. Application Development: What is MapReduce?
      “MapReduce is a programming model and an associated implementation for processing and generating large data sets.
      Programs written in this functional style are automatically parallelized and executed on a large cluster of commodity machines.
      This allows programmers withoutany experience with parallel and distributed systems to easily utilize the resources of a large distributed system.”
      Jeffrey Dean and Sanjay Ghemawat, Google, Inc.
      developer.amazonwebservices.com
      Copyright © BI Research 2009
    19. Developer Directions: A Changing Landscape
      “…the significance of Map-Reduce is in demonstrating the benefits of data-parallel programming to new classes of developers.This opens opportunities … by developing more powerful and efficient languages and runtime mechanisms that help these developers address more complex problems.”The Claremont Report on Database ResearchMay 28, 2008.http://db.cs.berkeley.edu/claremont/claremontreport08.pdf
      Copyright © BI Research 2009
    20. Hadoop vs RDBMS: A High-Level View
      Copyright © BI Research 2009
    21. The Many Dimensions of an Analytical Solution
      Functionality
      reportingreactive/predictive analysisperformance managementdata discovery/access/integrationdelivery/collaborationmetadata management
      Deployment model
      open sourceembeddedSaaSappliance
      Types of user
      BI consumerBI producerBI developerapplication developer
      Analytical solution
      Scalability & performance
      # of users# of data sourcesdata volumequery volumequery complexity
      Support
      product maintenance/supporteducationdocumentationconsulting services
      System functionality
      authentication/securityavailability/recoveryworkload managementoperations managementarchitecture/interfaces
      Copyright © BI Research 2009
    22. Using Hadoop in the Enterprise: Hadoop Developer
      HadoopMR applications
      HDFS
      queryaccess
      load(push)
      Hadoop connector
      operational data & content
      TraditionalETLapplications
      SQLanalyticalapplications
      analyticalRDBMS
      (Note that the HADOOP MR applications may run on a separate or the RDBMS platform)
      Copyright © BI Research 2009
    23. Using Hadoop in the Enterprise: SQL Developer
      HadoopMR applications
      HDFS
      Hadoop connector
      load(pull)
      operational data & content
      SQL/MRETLapplications
      SQL/MRanalyticalapplications
      analytical RDBMS +MR functions
      Copyright © BI Research 2009
    24. Summary
      The industry is moving toward providing a number of solutions for handling big data
      Hadoop and analytical RDBMSs are examples of this trend
      Both Hadoop and analytical RDBMSs have their strengths, weaknesses and uses
      Hadoop is useful for the processing of large data files with high availability where the structure of the data is ill-defined
      Analytical RDBMSs are well suited to analyzing complex data systems with high availability and integrity
      Hadoop and RDBMS applications can coexist and interoperate with each other
      Several RDBMS vendors provide Hadoop interoperability features
      Copyright © BI Research 2009
    25. Topics
      Aster Introduction
      Analyst Perspective: Colin White on Hadoop
      Introducing Aster-Hadoop Data Connector
      Case Studies: LinkedIn, ShareThis, comScore
      Summary
      25
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
    26. Aster In-Database MapReduce
      26
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
    27. Advantages of In-Database MapReduce
      27
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
      Evolves from Google’s well-known MapReduce framework
      Ultra-Fast Parallelism: Executes right where the data resides
      Advantages:
      • Leverages SQL as the base for connecting applications
      • Rich portfolio of supported languages – Java, Python, Ruby, Perl, C++, R, etc
      • Expressive flexibility (easily turing complete)
      • Unique reusability across applications (polymorphism)
      • High availability and resource management (sandbox, WLM)
      Power of Aster’s SQL/MR
      Expressive flexibility(language independent)
      Move execution file into Aster’s data store
      Reuse – by analysts or BI tools
      Write a SQL/MR function using Perl, Java, Python…(One-time)
      Upload the scriptor .jar file to Aster(One-time)
      Invoke SQL/MRfunction through SQL(Infinite times)
    28. Introducing Aster Hadoop Connector
      Why Is It Needed?
      Hadoop can be used for “munging” large unstructured data
      Challenge: slow, tedious manual operations required to transfer data from Hadoop into Aster nCluster
      What Is It?
      A set of 2 SQL/MR functions developed by Aster Data
      LoadFromHadoop: Parallel data loading from HDFS to Aster nCluster
      LoadToHadoop: Parallel data loading from Aster nCluster to HDFS
      Advantages: Parallel performance, Seamless (SQL), Consistency (ACID)
      28
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
      Example:
      insert into mytable
      select *
      from
      load_from_hadoop(
      on mytable
      host(&apos;10.10.3.22&apos;)
      port(9000)
      delimiter(&apos;,&apos;)
      nullstring(&apos;&apos;) files(&apos;hdfs_input_filepaths.txt&apos;)
      );
    29. Reports, Analytics, Applications
      (SQL / ODBC / JDBC)
      Queries
      nCluster + Hadoop: Complementary Use Case
      Data Sources
      (Unstructured Data)
      Queen
      Workers
      Loaders/Exporters
      Hadoop ETL
      Aster nCluster Database
      5
      1
      SQL/MR
      4
      2
      Hadoop Connector
      Intuitive, seamless Hadoop Connector for ultra-scalable parallel data transfer!
      3
      1. Raw, unstructured data loaded into Hadoop
      2. Unstructured data transformation
      3. Transformed data loaded into nCluster (directly to workers)
      SQL/MR functions invoke HDFS Connector
      4. SQL/MR processing of data (optional) and redistribution
      5. Analytic queries consume data received from Hadoop
      29
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
    30. Topics
      Aster Introduction
      Analyst Perspective: Colin White on Hadoop
      Introducing Aster-Hadoop Data Connector
      Case Studies: LinkedIn, ShareThis, comScore
      Summary
      30
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
    31. Case Study: HDFS-to-Aster
      Objective: Fast data transfer from Hadoop (HDFS) to Aster’s MPP Datawarehouse for purposes of faster, deeper data analysis
      After Aster: Ultra-fast, seamless HDFS to Aster data copy
      • HDFS to Aster SQL/MR function: 100 lines code developed in 2 days
      • Seamlessly invoked from SQL for ultra-fast parallelized HDFS reads
      • Latency cut down to 1.5 hours, a 10x performance improvement
      Before Aster: Tedious, high-latency loading
      • Sequential batch inserts via multi-step, multi-tool approach
      • Required 15 hours to load; gave up after 2 hours (unacceptable SLAs)
      31
      31
      Confidential and Proprietary. Copyright © 2009 Aster Data Systems
    32. Company Overview:
      ShareThis.com is the sharing network which makes it simple to share any online content
      Challenges faced in large-scale analytics, given existing options:
      • Difficult learning curve for data analysts (eg. statisticians/scientists) to gain sufficient expertise with Hadoop operations to use it for large-scale MapReduce problems
      • Hadoop as an open source project experiences frequent churn on features and API, forcing revisions to application code – i.e., much distraction from algorithm development
      “At ShareThis, we use a mix of Aster SQL/MR and Hadoop. But while developers have the background to dive into the Hadoop API, there is a steep learning curve for a larger class of data analysts where it’s just not practical to debug Hadoop programs. For those with less development experience, SQL/MR offers immediate accessibility. It only takes minutes to go from an idea to a prototype using Aster’s SQL/MR.”
      -- Paco Nathan, Principal Scientist, ShareThis
      ShareThis – Uses SQL/MR in the Cloud
    33. comScore, Inc.
      Founded in 1999
      Publically traded on NASDAQ (SCOR)
      Acquired MediaMetrix in 2002, and M:Metrics in 2007
      Corporate headquarters: Reston, VA
      Offices in Chicago, NYC, San Francisco, Seattle, Toronto, London, Tokyo, and Paris
      500+ full-time employees
      Experienced senior leadership team with a unique record of innovation in the market research industry
      More than 1200 clients across many industries
    34. Advising Hundreds of Leading Businesses(partial list)
      Internet
      Agencies
      Telecom
      Financial
      Retail
      Travel
      CPG
      Pharma
      Technology
    35. Disclosures
      We are in the building stage and are new to Hadoop and Aster Data
      We are working with best groups in the business:
      Aster Data
      Cloudera
      Impetus Technologies
      Partnering with Akamai on Edge
      We are under multiple NDAs with various vendors
    36. Goals
      Be able to scale to support an initial monthly volume of 160 Billion records
      Store 3 months of data online
      Be able to add incrementally to the environment to support growth
      Support advanced analytics
      Support end user access to record level data, preferably through a SQL interface
      Support the storage of row level data
    37. POC Analysis - Aster nCluster
      Built out small nCluster environment
      1 Queen; 3 Workers
      650+GB total storage
      Compared timings vs. NGBA array
      Findings
      Aster nCluster 3 – 10 times faster
      Plans
      Scale to 1 Queen, 24 Workers and 5 Loaders
      192TB total storage
      Focus on Analysts rather than Developers
    38. POC Analysis - Hadoop
      Built out Hadoop cluster on 0.18 on 6 servers
      Ported heavy C++ agg process to Java Map / Reduce
      Current analysis tasks
      Identified issue with memory allocation with guidance from Cloudera
      Modifying code to use persistent shared memory
      Findings
      Need to have more local RAM (originally had 8GB, need at least 16GB)
      Plans
      Use for batch aggregation processing
      Focus on Developers rather than Analysts
      Larger cluster size
    39. Topics
      Aster Introduction
      Analyst Perspective: Colin White on Hadoop
      Introducing Aster-Hadoop Data Connector
      Case Studies: LinkedIn, ShareThis, comScore
      Summary
      39
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
    40. Summary Takeaways
      40
      Confidential and proprietary. Copyright © 2009 Aster Data Systems
      Powerful, easy analytics on full data set with SQL/MR
      Plus new Hadoop Connector for seamless, parallel data loading!
      High-performance & massively scalable
      Low-cost incremental scaling on commodity hardware
      Fully-automated (one-click) administration
      Best Young Tech
      Entrepreneurs 2009
    41. Learn more
      www.asterdata.com
      Contact us
      hello@asterdata.com
      Questions and Answers
      41
      Confidential and proprietary. Copyright © 2009 Aster Data Systems

    + AsterDataAsterData, 1 month ago

    custom

    1804 views, 13 favs, 3 embeds more stats

    Colin White and discusses Hadoop and data warehousi more

    More info about this document

    © All Rights Reserved

    Go to text version

    • Total Views 1804
      • 1564 on SlideShare
      • 240 from embeds
    • Comments 0
    • Favorites 13
    • Downloads 0
    Most viewed embeds
    • 216 views on http://www.asterdata.com
    • 23 views on http://www.biblogs.com
    • 1 views on http://www.iweb34.com

    more

    All embeds
    • 216 views on http://www.asterdata.com
    • 23 views on http://www.biblogs.com
    • 1 views on http://www.iweb34.com

    less

    Flagged as inappropriate Flag as inappropriate
    Flag as inappropriate

    Select your reason for flagging this presentation as inappropriate. If needed, use the feedback form to let us know more details.

    Cancel
    File a copyright complaint
    Having problems? Go to our helpdesk?

    Categories