5. Evolution of Intel Pentium Pentium I Pentium II Chip area breakdown Pentium III Pentium IV Q: What can you observe? Why? ref: Zhenyu Ye / Bart Mesman / Henk Corporaal “GPU Architecture and Programing”
6. Extrapolation of Single Core CPU If we extrapolate the trend, in a few generations, Pentium will look like: Of course, we know it did not happen. Q: What happened instead? Why?
7. Evolution of Multi-core CPUs Penryn Chip area breakdown Bloomfield Gulftown Beckton Q: What can you observe? Why?
8. The Brick Wall -- UC Berkeley's View Power Wall: power expensive, transistors free Memory Wall: Memory slow, multiplies fastILP Wall: diminishing returns on more ILP HW Power Wall + Memory Wall + ILP Wall = Brick Wall David Patterson, "Computer Architecture is Back - The Berkeley View of the Parallel Computing Research Landscape", Stanford EE Computer Systems Colloquium, Jan 2007, link
10. Xpress disk (xpd) 1 TB: 20 GB/s IOPS: 600,000 reads 500,000 writes The NVIDIA Tesla c2070 memory 6GB interface 384bit 144 GB/s Fusion ioDrive Octal Capacity 5.12TB 6GB/s Read 4GB/s Write 1,000,000 IOPS 8K Camera: 260 km by a fiber optic network 3 GB/s a 20 minute broadcast would require roughly 4 TB of storage
11. RDBMS in MPP Architecture Useful work of a RDBMS Each SMB 6.4 GT/s
12. “Typical” Business Intelligence Today Query Query Slow Painful Expensive Slow Painful Expensive ETL ETL Copy Copy Reporting Data Warehouse Business Applications Calculation Engine Business Intelligence Query Results Data Warehouse Business Applications Calculation Engine Business Intelligence Query Results DataMarts Data Data Aggregates Aggregates Indexes Indexes Operational Data Store Operational Data Store
13. In-Memory Computing Disk is 1Mx slower than direct memory: like a chef doing his shopping on mars
14. In-Memory Computing Costs have Plummeted Cost of 1 Mb of memory in 2000: ≈$1 Bosphorus bridge: 105m / 344ft
15. In-Memory Computing Costs have Plummeted Cost of 1 Mb of memory today: <1 cent And shrinking…. Dell: Quad 10 Core server 512GB Ram, Hosting Services $5,200/month Child: 1.5m
20. Operations and Analytics Together Add ACID-compliant, row-based, in-memory Single source of data Faster, better BI and actionable intelligence Faster, better applications New application opportunities Copy Business Applications Business Intelligence Data Analytic Appliance
22. In-Memory Computing – The Time is NOW HW Technology Innovations SAP SW Technology Innovations Row and Column Store Multi-Core Architecture (8 x 8core CPU per blade) Massive parallel scaling with many blades One blade ~$50.000 = 1 Enterprise Class Server Compression 64bit address space – 2TB in current servers 100GB/s data throughput Dramatic decline in price/performance
23.
24. A simple aggregate can be processed in one linear scan mapping to memory organize by row A 10 € B 35 $ C 2 € D 40 € E 12 $ organize by column A B C D E 10 35 2 40 12 € $ € € $ memory address
25. In-Memory Computing – The Time is NOW HW Technology Innovations SAP SW Technology Innovations Row and Column Store Multi-Core Architecture (8 x 8core CPU per blade) Massive parallel scaling with many blades One blade ~$50.000 = 1 Enterprise Class Server Compression Partitioning 64bit address space – 2TB in current servers 100GB/s data throughput Dramatic decline in price/performance No Aggregate Tables
31. Foundation for new category of applications (e.g., planning, simulation) to significantly outperform current applications in category
32. Minimize data duplicationSAP In-Memory Appliance (SAP HANA™) SAP BusinessObjects tools Other query tools SQL MDX BICS SQL SAP HANA SAP In-Memory Computing Studio SAP In-Memory Database Row & Column Storage Calculation and Planning Engine Real-Time Data Replication SAP Business Objects Data Services SAP Business Suite Other data sources SAP NetWeaver Business Warehouse
33. HANA Combines Software and Hardware In-Memory Computing Engine (Software) + Pre-Installed Systems (Hardware)
34. S+ S XS HANA Appliance “T-shirt” sizes Specifications & Approximate Data Volumes Starts at S and scales up to M
35. L M M+ HANA Appliance “T-shirt” sizes Specifications & Approximate Data Volumes Starts at M and scales up to L
39. The First 35 Years: Innovated with ERP & LOB Apps Three Years Ago: Innovated with Analytics Last Year: Innovated with Mobility This Year: Innovating the Database HANA Accelerates Data, Applications, Analytics Long Term: HANA Is the Database Mobility Accessible Systems ERP + LOB Business Analytics Data “In” BICS Info “Out” Systems of Record Systems of Engagement Business Applications Performance Bound by Data Oracle DB2 SQL Other HANA In Memory Database HANA In Memory Database Oracle SQL DB2, etc. ELT or ETL ELT or ETL
40. The Truth About Enterprise Data Landscapes The Value of HANA The Future of “The Stack” – HANA Nirvana ? Queries Ad-Hoc Reports ETL DATA QUALITY Dashboard Information Semantic Universe OLAP OLAP OLAP OLAP Dimensional HANA HANA Mart Mart Mart Mart Mart Mart Mart Marts BW TeraData Netezza IQ BW/Netezza/Teradata/IQ Warehouses Oracle Oracle/DB2/SQL/Other DB2 Other SQL Operational Applications
41. SAP BusnessObjects Event Insight Event InsightNode Event Insight Node Event InsightNode WWW Supplier Mfg Plant Apps Apps Event InsightNode Event Insight Apps Custom DB Event InsightNode DSD DistributionCenter Retailer A Distribution Center Distribution Center Event InsightNode Event InsightNode Apps BW Retail Store A Retail Store B
42. SAP BusnessObjects Event Insight Event InsightNode Event InsightNode Event InsightNode WWW Event Pattern Supplier Event Pattern Identified Define Event Pattern Mfg Plant Apps Apps Event InsightNode Event InsightNode Apps Custom DB Event InsightNode DSD DistributionCenter Retailer A Distribution Center Distribution Center Event Insight Node Event Insight Node Apps BW Retail Store A Retail Store B
Editor's Notes
Memory speeded up exponentially. But disk access didn’t— it’s only 12.5x faster than half a century ago (from 1,200 revolutions per minute in 1956 to 15,000 RPM today), because of the limitations of aerodynamics: disks would literally fly off the spindle.The result is that it can be a million times quicker to access data in memory than off disk. That’s like having to get in a space ship to another planet in order to buy the ingredients you need for a meal. But the problem was that memory was incredibly expensive — like having to pay millions of dollars for every square inch of storage in the kitchen. Over the years, it got cheaper to have more storage, and so database designers were able to do the equivalent of storing the most-often used foods locally, more people fetching food, precooking some of the food in advance, etc. But the raw food itself was still stored in the only viable place — that huge, distant warehouse.
Turns out that most standard business applications — e.g. finance — are sparse, require sophisticated calculations (such as budget allocations) , and need auditing — run particularly well on in-memory column databases….
SAP was founded on an innovative idea – to create standardized business software. Solutions such as ERP and line of business applications form the core for thousands of the world’s best run companies today. <click>We innovated again three years ago by extending that core with integrated analytics. Now you could not only run your core business processes efficiently you could also manage, govern and gain insight into your business. <click>And we innovated again by extending our portfolio to mobile devices, making your business accessible – not only in more places, but to millions more people.Operations, management and accessibility are all very valuable. What could we do next which would be equally valuable? <click>We recognized an opportunity to address significant issues that businesses face in the area of performance – by tackling the data layer itself. The performance of every application in the SAP portfolio depends on the underlying data management architecture. It even impacts the way we design and code our software. <click>So, the next innovation is HANA. Using the latest developments in high performance server computing, combined with innovative data management software we are able to significantly accelerate data access, simplify applications, and transform the way analytics processes work. HANA delivers speed and simplicity. It accelerates business applications and reduces the cost of your IT landscape. That is the fundamental value of this technology – It makes everything we have done before even better.<click>Our ultimate goal is to combine transactional and analytical data management using HANA. It is technically possible, so why wouldn’t we leverage that speed and simplicity in everything we do? Why wouldn’t we deliver as much value to you, our customer, as we can?
Let’s take a more detailed look at that complexity.<click>There are a lot of layers in between those operational data stores and the reports and dashboards needed to run your business. There are warehouses and marts and cubes and universes, all governed by copying and quality management processes. <click>Data moves up through these layers, aggregating and transforming to become the ‘right information’ at the ‘right level of detail’. The more systems you have at the bottom (supply), and the more requests you have at the top (demand), the more complex the middle is going to be – your internal ‘data economy’ so to speak.<click>What HANA does is to cut out the middle-men in that economy. It synchronizes transactional information in real time. It processes data at amazing speed. It can pre-aggregate information for you. And it can handle new queries on-the-fly, so when the business has a new request IT does not have to rely on middle-men to fulfill it. <click>How far can we take that idea? There is no reason why a traditional, disk-based system should be needed at all. So, it is our goal as a company to take them out of the picture.