Microsoft SQL Server, Scalability, & Database Research Jim Gray Researcher Microsoft Corporation Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 1 3 October 1999 Chicago, Ill. Outline Summary of what you heard. (10 min) The database scene in general. (10 min) Scaleability: Farms, Clones,Parts & Packs (15 min) Microsoft DB research focus. (15 min) • TerraServer (design and ops). • RAGS. • Data Mining Q&A Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ (10 min) 2 3 October 1999 Chicago, Ill. Organizations Are Going Online Building a digital nervous system. Inexpensive hardware means huge databases are possible. But, we are drowning in data. Databases help organize information. Microsoft’s goal: • Information at your fingertips. • Make it easy to capture, manage, and analyze information. Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 3 3 October 1999 Chicago, Ill. Microsoft SQL Server 7 Goals Easy Scalability Data Warehousing Dynamic self management Multi-site management Operation Scripting Job scheduling and execution Alert/response management Scriptable Install+upgrade DBA profiling/tuning tools Unicode English Language Query Integrated with NT Security Integrated with NT files Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 4 3 October 1999 Chicago, Ill. Scalability Scalability Data Warehousing Easy Win9x/NTW version Dynamic row-level locking Improved query optimizer Intra-query parallelism VLDB improvements Replication improvements Distributed query High Availability Clusters Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 11 3 October 1999 Chicago, Ill. Scale Down to Windows 95-98 Full function (same as NTW) Integration with Access 97 MSDE in Office2000 and MSDN WinCE version demonstrated Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 12 3 October 1999 Chicago, Ill. Replication Publisher Transactional and Merge Remote update ODBC and OLE DB subscribers Wizards Performance OS 390 DB2 2PC, RPC Distributor DB2 VSAM Subscriber CICS Subscriber Subscriber Subscriber Updating Subscriber (immediate updates) Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 13 3 October 1999 Chicago, Ill. Query Processor Enhancements Focus on Complex Queries Parallelism Improved scan, fetch, & sort Smart hash & merge join Large joins & grouping Better query optimization Multi-index operations Automatic statistics maintenance Distributed Query Heterogeneous Query Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 14 3 October 1999 Chicago, Ill. Parallel Query SMP & Disk Parallelism Global Agg. Result 50 rows + 4 x 50 rows Local Agg. + + + + 50,000 rows •# of emp. per group Disks •total inc. per group Plus Distributed Plus Hash Join (fanciest on the planet) Plus Optimized Partitioned views Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 15 3 October 1999 Chicago, Ill. Distributed Heterogeneous Queries Data Fusion / Integration Join spread sheets, databases, directories, Text DBs etc. Any source that exposes OLE DB interfaces SQL Server as gateway, even on the desktop Directory Service Database (DB2, VSAM, Oracle, …) Spreadsheet SQL 7.0 Query Processor Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ Photos Mail Maps Documents and the Web 3 October 1999 16 Chicago, Ill. Utilities The Key to LARGE Databases Auto-Repair Index creation Backup • Fuzzy • Parallel • Incremental • Restartable ~2x faster than 6.5 DBCC • not required, • a good practice • 5x - 100x faster Recovery • Fast • File granularity • shrinks file • reclusters file Recovery time (secs) Reorganize 60 50 SQL Server 6.5 SQL Server 7.0 40 30 20 10 0 Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 1 2 3 # of indices 17 4 3 October 1999 Chicago, Ill. Data Warehousing Warehousing Framework Visual data modeler Microsoft repository Data transformation services Scalability (DTS) Plato & Dcube - Multi Dimensional Data Cubes Data Warehousing English query 2.0 Easy Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 20 3 October 1999 Chicago, Ill. Data Warehouse / Data Analysis Data Transformation Services to get data into the warehouse CUBE (OLE/DB OLAP) to analyze data Operational Data Extact & Load OLAP Data Warehouse Storage Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 22 3 October 1999 Chicago, Ill. Plato and Data Cube and HOLAP By Year By Make By Make & Year Source table Europe RED WHITE BLUE By Color & Year Sum Partition 1 By Color “Plato” User 1 ROLAP Dcube SQL Designer USA Partition 2 MD SQL Client app “Plato” server Asia Dcube Partition 3 User 2 ROLAP Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ Client app 25 3 October 1999 Chicago, Ill. English Query Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 26 3 October 1999 Chicago, Ill. Easy Scalable Data Warehousing Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 27 3 October 1999 Chicago, Ill. “Shiloh” The Next SQL Server Shiloh (H1’00) - Strengthen Position • Data Warehousing leadership Materialized Views Cascading Referential Integrity (#1 requested user-group feature) XML support • Scalability WinCE support W2K VLM (36 and 64 bit) Multi-instance support Yukon – Next Big Step • Scalability (Clusters, Partitions) • Programmability • Ease of Use (Self Tuning, Auto Config) Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 3 October 1999 Chicago, Ill. Outline Summary of what you heard. (10 min) The database scene in general. (10 min) Scaleability: Farms, Clones,Parts & Packs (15 min) Microsoft DB research focus. (15 min) • TerraServer (design and ops). • RAGS. • Data Mining Q&A Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ (10 min) 29 3 October 1999 Chicago, Ill. Info Capture Yotta Zetta Exa You can record everything you see or hear or read. What will you do with it? How will you organize & analyze it? Most data will never be seen Analysis an summarization are key technologies Everything ! Recorded All Books MultiMedia Peta Tera Giga Mega Kilo All books (words) .Movi e A Photo Video Audio Read or write: 8 PB per lifetime (10GBph) 30 TB (10KBps) 8 GB (words) See: http://www.lesk.com/mlesk/ksg97/ksg.html A Book Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 30 3 October 1999 Chicago, Ill. Data Tidal Wave Seagate 47GB drive @ 783$ (= 1.7 ¢/mb) • 100 GB penny per MB drive coming in 2000 10 $/GB = 10 k$/ Terabyte! • “Everyone” can afford one What’s a terror bite? • • • If you sell ten billion items a year (e.g Wal-Mart) And you record 100 bytes on each one Then you get a TeraByte/year Where will the terror bytes come from? • Multimedia (like the TerraServer) and... Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 31 3 October 1999 Chicago, Ill. Reducing Data’s Cost-of-Ownership Self-Managing data Cost of ownership: One admin/TB (100K$ vs 10K$) Admin cost exceeds storage cost. SQL 7: Suggests indices Migrates data away from end of file Truncates file Someday: Automatic move files to balance disks Online defragmentation & restructuring Online physical redesign Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 32 3 October 1999 Chicago, Ill. OBJECT RELATIONAL The Next Great DBMS Wave All DB vendors have added objects to DB Microsoft is adding DBs to Objects Integration with COM+ Gives user-defined types and objects Plug-ins will be Billion dollar industry • Blades for SQL Server razor Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 33 3 October 1999 Chicago, Ill. Why Is XML Important? Self-describing data Data stream in a typical interface… “ABC47-Z”, “100”, “STL”, “C”, “3”, “28” Same data stream in XML… <INVENTORY> <PART_NUM>ABC47-Z</PART_NUM> <QUANTITY>100</QUANTITY> <WAREHOUSE>STL</WAREHOUSE> <ZONE>C</ZONE> <AISLE>3</AISLE> <BIN>28</BIN> </INVENTORY> Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 34 3 October 1999 Chicago, Ill. table.xsl bar.xsl art.xsl Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 35 3 October 1999 Chicago, Ill. XML Applications Exposing Software as a “Service” • Websites without UI’s • Exposed services with common scheme • Integration points at the enterprise, valuechain, workgroup, desktop and intelligent gizmo “levels” B2B value chains • Uses XML to transmit wide range of date to a broad set of stakeholders (regulatory agencies, suppliers, customers, etc.). • Leverage for prior efforts like EDI • BizTalk a key industry effort in this regard Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 36 3 October 1999 Chicago, Ill. XML: BizTalk Framework www.biztalk.org XML XML schema XML XML MVS CICS SAP R/3 Library Order Processing Service Interface XML XML XML Message Another Service XML Document XML Message XML Document JD Edwards Browser Client Apps New Form Factors Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 37 3 October 1999 Chicago, Ill. Outline Summary of what you heard. (20 min) The database scene in general. (10 min) Scaleability: Farms, Clones,Parts & Packs (10 min) Microsoft DB research focus. (15 min) • TerraServer (design and ops). • RAGS. • Data Mining Q&A Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ (15 min) 38 3 October 1999 Chicago, Ill. Terminology for scaleability Farm Farms of servers: • Clones: identical Scaleability + availability Partition Clone • Partitions: Scaleability • Packs Pack Partition availability via fail-over Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 39 3 October 1999 Chicago, Ill. Unpredictable Growth The TerraServer Story: • • • • We expected 5 M hits per day We got 50 M hits on day 1 We peak at 15-20 M hpd on a “hot” day Average 5 M hpd after 1 year Most of us cannot predict demand • Must be able to deal with NO demand • Must be able to deal with HUGE demand Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 40 3 October 1999 Chicago, Ill. An Architecture for Internet Services? Need to be able to add capacity • New processing • New storage • New networking Need continuous service • Online change of all components (hardware and software) • Multiple service sites • Multiple network providers Need great development tools • Change the application several times per year. • Add new services several times per year. Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 41 3 October 1999 Chicago, Ill. Premise: Each Site is a Farm Buy computing by the slice (brick): • Rack of servers + disks. Grow by adding slices • Spread data and computation to new slices Two growth styles: 1997 Microsoft.Com Farm Building 11 Staging Servers (7) Internal WWW FTP Servers Download Replication Router SQL Reporting Live SQL Server www.microsoft.com (4) premium.microsoft.com European Data Center www.microsoft.com (1) (3) SQLNet SQL SERVERS Feeder LAN SQL Consolidators (2) RouterDMZ Staging Servers FTP Live SQL Servers Download Server (1) MOSWest Switched Admin LAN Ethernet MOSWest register.microsoft.com www.microsoft.com msid.msn.com (1) (2) (4) search.microsoft.com (3) home.microsoft.com home.microsoft.com FDDI Ring (3) (4) (MIS2) premium.microsoft.com (2) activex.microsoft.com (2) FDDI Ring (MIS1) cdm.microsoft.com (1) Router Router msid.msn.com Router (1) Primary Router Gigaswitch premium.microsoft.com Router www.microsoft.com (1) Router (3) Secondary Gigaswitch Router Router FTP.microsoft.com (3) FDDI Ring home.microsoft.com(MIS3) msid.msn.com www.microsoft.com (2) (1) (5) 0 register.microsoft.com FDDI Ring (2) (MIS4) register.microsoft.com home.microsoft.com support.microsoft.com (1) (5) register.msn.com (2) (2) support.microsoft.com search.microsoft.com(1) (3) search.microsoft.com (1) msid.msn.com (1) Router Japan Data Center www.microsoft.com SQL SERVERS premium.microsoft.com(3) (2) (1) msid.msn.com (1) Switched Ethernet FTP Download Server (1) HTTP search.microsoft.com Download Servers (2) (2) Router 2 OC3 (45Mb/Sec Each) 2 Ethernet (100 Mb/Sec Each) 13 DS3 (45 Mb/Sec Each) Internet • Clones: anonymous servers • Parts+Packs: Partitions fail over within a pack In both cases, remote farm for disaster recovery Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 42 3 October 1999 Chicago, Ill. Scaleable Systems Scale UP and Scale OUT Everyone does both. Choice is • Size of a brick • Clones or partitions • Size of a pack Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 43 3 October 1999 Chicago, Ill. Everyone scales out What’s the Brick? 1M$/slice • IBM S390? • Sun E 10,000? 100 K$/slice • Wintel 8X 10 K$/slice • Wintel 4x 1 K$/slice • Wintel 1x Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 44 3 October 1999 Chicago, Ill. Clones: Availability+Scalability Some applications are • Read-mostly • Low consistency requirements • Modest storage requirement (less than 1TB) Examples: • HTTP web servers (IP sprayer/sieve + replication) • LDAP servers (replication via gossip) • App/compute servers or firewalls Replicate app at all nodes (clones) Spray requests across nodes. Grow by adding clones Fault tolerance: stop sending to dead clone. Growth: add a clone. Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 45 3 October 1999 Chicago, Ill. Facilities Clones Need Automatic replication • Applications (and system software) • Data Automatic request routing • Spray or sieve Management: • Who is up? • Update management & propagation • Application monitoring. Clones are very easy to manage: • Rule of thumb: 100’s of clones per admin Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 46 3 October 1999 Chicago, Ill. Partitions for Scalability Clones are not appropriate for some apps. • Statefull apps do not replicate well • high update rates do not replicate well • Huge DBs (disk to expensive to clone) Examples • Email / chat / … • Databases Partition state among servers Scalability (online): • Partition split/merge • Partitioning must be transparent to client. Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 47 3 October 1999 Chicago, Ill. Partitioned (aka. Clustered) Apps Mail servers • Perfectly partitionable Business Object Servers • Partition by set of objects. Parallel Databases Transparent access to partitioned tables Parallel Query Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 48 3 October 1999 Chicago, Ill. Packs for Availability Each partition may fail (independent of others) Partitions migrate to new node via fail-over • Fail-over in seconds Pack: the nodes supporting a partition • • • • • VMS Cluster Tandem Process Pair SP2 HACMP Sysplex™ WinNT MSCS (wolfpack) Cluster In A Box now commodity Partitions grow in packs. Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 49 3 October 1999 Chicago, Ill. What Parts+Packs Need Automatic partitioning (in dbms, mail, files,…) • Location transparent • Partition split/merge • Grow without limits (100x10TB) Simple failover model • Partition migration is transparent • MSCS-like model for services Application-centric request routing Management: • Who is up? • Automatic partition management (split/merge) Gray, Research and Microsoft SQL Server • Jim Application monitoring. 50 Microsoft Research http://research.Microsoft.com/~gray/talks/ 3 October 1999 Chicago, Ill. Services on Clones & Partitions Application provides a set of services If cloned: • Services are on subset of clones If partitioned: • Services run at each partition System load balancing routes request to • Any clone • Correct partition. • Routes around failures. Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 51 3 October 1999 Chicago, Ill. Farm pairs: Always Up Two farms Changes from one sent to other When one farm fails other provides service Masks • Hardware/Software faults • Operations tasks (reorganize, upgrade move • Environmental faults (power fail) Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 52 3 October 1999 Chicago, Ill. Availabilty for a simple web site Clones for availability Packs for availability Web File Store SQL Database SQL Temp State Front End Load Balance Web Clients Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 53 3 October 1999 Chicago, Ill. Farm Scale Out Scenarios The FARM: Clones and Packs of Partitions Packed Partitions: Database Transparency SQL Partition 3 SQL Partition 2 replication Web File StoreA Web File StoreB SQL SQLPartition1 Database SQL Temp State Cloned Packed file servers Web Clients Load Balance Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 54 Cloned Front Ends (firewall, sprayer, 3 October 1999 web server )Chicago, Ill. Reliable, Scalable, Modular Network Load Clients Balancing Clones 1 Component Load Balancing (COM+) Clones Cluster Service Pack 1 2 2 3 4 3 … … 32 8 COM+ Components IIS Web Server Application Servers Gray, Research and services Microsoft SQL Server orJimother IP based Microsoft Research http://research.Microsoft.com/~gray/talks/ 55 Data Servers 3 October 1999 SQL, Exchange, File Chicago, Ill. Talk 2 (if there is time) Terminology for scaleability Farms of servers: Farm • Clones: identical Scaleability + availability Partition Clone • Partitions: Scaleability • Packs Pack Partition availability via fail-over Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 56 3 October 1999 Chicago, Ill. Scalability: COM+ progress serving 1,000-statement ASP’s (servelets) Poor SMP Scaleability on IIS4 NT4 450 Big SPS: servelets per second improvements 400 (ASPs served per second by IIS,1P from standard 350 1,000 statement VBscript) 300 Transaction 250 Processing tricks 200 Shift from 4x200 Mhz to 8 450 Mhz Out of Proc 150 (safe execution) 100 now much faster 50 than In Proc 0 was on IIS4 NT4 W2K W2K W2K NT4 InProc B1 RC1 RC2 inProc InProc InProc Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ OOP 57 2P W2K B1 OOP 4P W2K RC1 OOP 8P W2K RC2 OOP 3 October 1999 Chicago, Ill. Scaleability: So, What about the death of NT/Alpha? Two simultaneous Compaq TPC-C numbers Intel Profusion Alpha NT/SQL/COM+ 550 Mhz 8 Processors 4 GB memory 40,368 TPM-C @ 18.46$/tpmC 745 K$ 5-year cost Avail: 12/31/99 Unix/Sybase/Tuexdo 700 Mhz 8 Processors 16 GB memory 42,437 TPM-C @ 55.45 $/tpmC $2.35 M$ 5-year cost Avail: 10/18/99 200% more expense for 5% more performance? Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 58 3 October 1999 Chicago, Ill. Outline Summary of what you heard. (10 min) The database scene in general. (10 min) Scaleability: Farms, Clones,Parts & Packs (15 min) Microsoft DB research focus. (15 min) • TerraServer (design and ops). • RAGS. • Data Mining Q&A Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ (10 min) 59 3 October 1999 Chicago, Ill. The TerraServer http://www.terraserver.microsoft.com/ Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 60 3 October 1999 Chicago, Ill. Database & application UI Coverage: Range from 70ºN to 70ºS today: 35% U.S., 1% outside U.S. Source Imagery: • Concept: User navigates an ‘almost seamless’ image of 4 TB 1sq meter/pixel Aerial (USGS - 60,000 earth 46Mb B&W- 151Mb Color IR files) • 1 TB 1.56 meter/pixel Satellite (Spin-2 - 2400 300 Mb B&W) Display Imagery: 200x200 pixel images, subsample to build image pyramid Store 5x compressed data Nav Tools: • • • • 1.5 m place names “Click-on” Coverage map Expedia & Virtual Globe Pick of the week Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 200x200 m tile .4 x.4 km browse .8 x .8 km 8m thumbnail 1.6x 1.6 km “city view” 61 3 October 1999 Chicago, Ill. Software: Classic 3 Tier Design Web Client Image Server Active Server Pages (ADO) 24 Internet Information Server 5.0 46 HTML Java Viewer browser MTS The Internet 20 (8/12) Terra-Server Stored Procedures Fire wall SQL Server 7 46 TerraServer DB TerraServer Web Site Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ Internet Information Server 4.0 Microsoft Site Server EE Image Delivery SQL Server Application 7 1999 Image Provider3 October Site(s) 62 Chicago, Ill. Logical Schema Load Mgmt Famous Category Country Name State Name Famous Place SourceMeta Scale Job Place Type Place Name Image Search Image Load Job Pyramid Small Place Name External Geo Image Type TerraServer Gazetteer External Link Imagery External Group Search Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ Terra Database TerraAdmin 63 3 October 1999 Admin Chicago, Ill. TerraServer File Group Layout Convert 324 disks to 28 RAID5 sets plus 28 spare drives Make 4 NT volumes (RAID 50) 595 GB per volume Build 30 20GB files on each volume DB is File Group of 120 files E: F: Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ H: G: 64 3 October 1999 Chicago, Ill. Hardware Internet Map Site Server Servers SPIN-2 100 Mbps Ethernet Switch DS3 Web Servers 2.9 TB Database Server AlphaServer 8400 8x400. 10 GB RAM 324 StorageWorks disks 10 drive tape library (STC Timber Wolf DLT7000 ) Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 65 3 October 1999 Chicago, Ill. Load & Backup&Recovery Backup and Recovery Performance • STK 9710 Tape robot Data Bytes Backed Up • SQL Server Backup & Total Time Number of Tapes Consumed Restore + Total Tape Drives • Legato Networker Data ThroughPut Average ThroughPut Per Device • Fast, incremental, Average Throughput Per Device differential, online NTFS Logical Volumes • Clocked at 80 MBps (peak) (~ 200 GB/hr) 1.2 TB 7.25 Hours 27 10 168 GB/Ho 16.8 GB/Hour 4.97 MB/Sec 2 Restore • Fast, incremental (file oriented), not online. Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 66 3 October 1999 Chicago, Ill. BAD OLD Load DLT Tape DLT Tape “tar” NT \Drop’N’ DoJob LoadMgr DB Wait 4 Load Backup LoadMgr ... LoadMgr ESA Alpha Server 4100 100mbit EtherSwitch 60 4.3 GB Drives Alpha Server 4100 ImgCutter \Drop’N’ \Images 10: ImgCutter 20: Partition 30: ThumbImg 40: BrowseImg 45: JumpImg 50: TileImg 55: Meta Data 60: Tile Meta 70: Img Meta 80: Update Place Enterprise Storage Array STC DLT Tape Library 108 9.1 GB Drives 108 9.1 GB Drives Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 108 9.1 GB Drives Alpha Server 8400 67 3 October 1999 Chicago, Ill. New Image Load and Update DLT Tape “tar” Active Server Pages Cut & Load Scheduling System Metadata Load DB Dither Image Pyramid From base Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ Image Cutter Merge ODBC Tx TerraLoader ODBC TX ODBC Tx 68 TerraServer SQL DBMS 3 October 1999 Chicago, Ill. After a Year: 30M Count 1 TB of data 750 M records 10M 2.3 billion Hits 2.0 billion DB Queries 0 1.7 billion Images sent 368 million Page Views 99.93% DB Availability 3rd design now Online Built and operated by team of 4 people TerraServer Daily Traffic Jun 22, 1998 thru June 22, 1999 Sessions Hit Page View DB Query Image 20M (Hours:minutes) 8640 6:00 7920 5:30 7200 5:00 6480 Operations 4:30 5760 4:00 5040 4320 3600 2880 Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ Down Time TotalTime (Hours) Up 3:30 3:00 2:30 Scheduled 2:00 2160 1:30 1440 1:00 720 0:30 0 0:00 70 HW+Software 3 October 1999 Chicago, Ill. What TerraServer Shows Can serve huge databases on Internet for about a penny a page view mostly phone bill (!). Advertising pays more than a penny a page. Commodity tools do scale fairly far. A few people (3 developers, 1 operator) using power tools can build an impressive web site Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 74 3 October 1999 Chicago, Ill. Outline Summary of what you heard. (20 min) The database scene in general. (10 min) Scaleability: Packs & Mobs (10 min) Microsoft DB research focus. (15 min) • TerraServer (design and ops). • RAGS. • Data Mining Q&A Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ (15 min) 75 3 October 1999 Chicago, Ill. Automatic Testing 60% of Microsoft R&D is testing. What can research do to help? • beyond joining the 500,000 Win2K beta testers Test generation robot: • Make up SQL queries • Send them to SQL Server, Oracle, DB2, Informix,… • If answer is the same, great, if not there is a problem Case W X Y Z 1672 1672 232 234 241 31 1 1 1 1 31 15 12 28 1 12 5 116 0 29 32 4 18 18 19 25 18 113 All four agree 84% 1672 1672 45 19 Error Also good for stress tests Found MANY bugs in our products (all fixed). Found MANY bugs in other’s products. Very valuable tool. W,X, and agree 95% Problem with intermediate table. MSR-TR-98-21 Massive Stochastic Testing of SQL, Slutz, Don http://research.microsoft.com/scripts/pubDB/pubsasp.asp?RecordID=175 Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 76 3 October 1999 Chicago, Ill. Kilo Mega Giga Tera Peta Some Tera-Byte Databases The Web: 1 TB of HTML TerraServer 1 TB of images Several other 1 TB (file) servers Hotmail: 20 TB of email Sloan Digital Sky Survey: 40 TB raw, 2 TB cooked EOS/DIS (picture of planet each week) • 15 PB by 2007 Exa Federal Clearing house: images of checks • 15 PB by 2006 (7 year history) Zetta Nuclear Stockpile Stewardship Program • 10 Exabytes (???!!) Yotta Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 77 3 October 1999 Chicago, Ill. Data Mining Find interesting structure (patterns, relationships) in data • Prediction • Segmentation (clustering) • Dependency modeling (find distribution) • Summarization • Trend and change detection and modeling Allow user to state the query in terms of the business logic • User does not speak statistics or SQL Use data to build predictors • regression, classification, segmentation etc. Generate summaries and reports for insight • find “easy to describe” segments in data automatically • find segments not known to analyst Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 78 3 October 1999 Chicago, Ill. Data Mining: Microsoft SiteServer Commerce 3.0 Http://www.holtoutlet.com/outlet4 Intelligent Cross-sell Based on: • Historical sales baskets in stores • Contents of current shopper basket • Browsing behavior of shopper Predict: ranking of products in store likely to be most interesting to shopper. Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 79 3 October 1999 Chicago, Ill. 100.0% 98.5% 94.8% 68.5% 56.9% 43.8% 34.5% 25.5% 6.7% 5.3% 1.3% 0.6% 0.3% 0.2% 0.1% 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% % Captured of true targets Mail to 25% and capture 40% 400% improved response! % mailed RealJimdata drawn from a Microsoft marketing example 3 October 1999 Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 80 Chicago, Ill. How do people use www.microsoft.com? 100M hits per day 14M users/week User browsing data X segments Data Mining (Clustering) Engine Cluster Visualizer Wizard Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 81 3 October 1999 Chicago, Ill. Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ 82 3 October 1999 Chicago, Ill. Outline Summary of what you heard. (10 min) The database scene in general. (10 min) Scaleability: Farms, Clones,Parts & Packs (15 min) Microsoft DB research focus. (15 min) • TerraServer (design and ops). • RAGS. • Data Mining Q&A Jim Gray, Research and Microsoft SQL Server Microsoft Research http://research.Microsoft.com/~gray/talks/ (10 min) 83 3 October 1999 Chicago, Ill.