SQL Server - Bài 13

SQL Server - Bài

 Overview of database performance and scalability

 Compatibility

 Advantages of 64-bit environment

 Consideration for choosing SQL Server 64-bit

 Deployment consideration

Database Performance and Scalability

 Many applications running on 32-bit platforms today are

approaching or have exceeded the limits of the platform

 Advanced DBMS capabilities, data is growing numbers of business users

 Organizations are consolidating servers to simplify critical data

center operations Consolidation can reduce management

complexity and cost, while reducing physical space requirements in the data center

SQL Server 64 bit

Cost Savings

 Great performance

 T-SQL code-compatibility with SQL Server 2000

 8 node clustering support

 Same on-disk format as 32-bit for easy migration

 One setup for database & OLAP based on Windows Installer technology

 Compelling alternative to expensive Unix solutions

Manageability

512 GB)

SQL Server 2000 – 64 bit running on an Itanium 2 Server

• Buffer pool pages can be mapped / unmapped

•As pages are

needed, they can

be mapped into the virtual

address space

•Some structure

stay in virtual address memory including plan cache, cursor structures and user connection context

•Cost associated

with mapping / unmapping

• Virtual address space available to all

•Current and future

hardware

The primary differences between the SQL 64-bit and SQL 32-bit

 The 64-bit architecture offers a larger directly-addressable memory space SQL

Server 2005 (64-bit) is not bound by the memory limits of 32-bit systems Therefore, more memory is available for performing complex queries and supporting essential database operations

 The 64-bit processor provides enhanced parallelism, thereby providing more linear scalability and support for up to 64 processors, and yielding stronger returns per

processor as compared to 32-bit systems.

 The improved bus architecture enhances performance by moving more data between cache and processors in shorter periods.

 A larger on-die cache allows for faster completion of user requests and more efficient use of processor time.

Larger numbers of processors and more

linear scalability per processor

 Improvements in parallel processing and bus architectures enable 64-bit platforms to support larger numbers of processors (up to 64) while providing close to linear scalability with each additional processor With a larger

number of processors, SQL Server can support more processes,

applications, and users in a single system

 Server platforms that offer more than 32 CPUs are available exclusively on 64-bit architecture The highest TPC-C benchmark figures for SQL Server have been achieved on 64-bit systems because these systems leverage both large amounts of memory and the superior scaling of 64-bit processors

on the 64-bit architecture.

Enhanced bus architecture

 The bus architecture on current 64-bit chipsets is faster and wider than earlier generations More data is passed

to the cache and processor; this is somewhat analogous

to the improvement that broadband connections offer

over dial-up connections.

Industry-Leading Features

 Analysis Server – Gartner claimed Microsoft Analysis Server best of breed

 Auto Configuration, Tuning

 Loading Data – Bulk Insert rates of 70 MB/Sec+

 Replication – transactional replication – 2000 cmds/sec

 Management Tools – Set the bar

 Reliability testing unmatched in the industry

 Utilities – Fast Backup, Best re-org and very fast Load

 SQL CE, Query Processor, Storage Engine, Developer talent, XML, etc.

Example Terabyte-Scale Deployments

 Large-Scale OLTP and Data Warehousing

 Call Center Mgmt / Telecom Billing – Verizon (19 TB)

 Credit Scoring and Collections – Fair Isaac (13 TB)

 Retail Analytics / BI – Edgars Consolidated (4 TB)

 Retail Banking Analytics – Royal Bank of Scotland (3 TB)

 SAP – Pilgrim’s Pride (3 TB)

 Brokerage Front and Back-Office – CSS Software (1+ TB)

 Loss Prevention: (US Retail Chain) – Rite Aid (1+ TB)

 Scientific / New Generation

 Bioinformatics: Rosetta Genomics – (10 TB)

 GeoSpatial: Terraserver – (3+ TB)

 Astronomy: SkyServer – (~1TB  5TB)

What is 64-bit SQL Server?

 Same codebase as SQL Server

 All pointer types are now 64 bit

 Good news: Lots of virtual memory

 Bad news: Control structures bigger

 Flat memory addressing to 512GB of RAM

 Does NOT require AWE support, massive cache useable by all components of

SQL: Stored proc, OLAP, sort area… etc.

 Massive Scale-up support

 64-way HP, 32-way NEC, 16-way Unisys

 Until very recently MSFT held fastest TPC number

 Easy Database Migration & Integration

 Detach/Attach, Log shipping, Replication etc.

 8-node Fail-over clustering support

 Maintains T-SQL code-compatibility with SQL Server 32-bit releases

 Same SQL Server 2000 code base plus service pack fixes and

performance enhancements

 No changes in database file format

 No changes in client applications when connecting to a 64-bit

backend versus a 32-bit backend

 Supported Clients

 SQL Server 2005 (64-bit) is optimized to run on servers using Intel

Itanium 2, AMD Opteron, AMD Athlon 64, Intel Xeon with EM64T support, or Intel Pentium IV with EM64T support processors

64-bit gaps

 Some non-engine SQL components are not currently

supported on 64-bit:

 Graphical Admin Tools

 Data Transformation Services

64-Bit Winning Scenarios

 Workloads stressing Virtual Memory

 Limited to 3GB on 32-bit, regardless of AWE

 Applications that stress

 Procedure cache and cursor memory

 Sort, index and hash table memory

64-Bit Winning Scenarios - 2

 Massive Scale-up Scenarios

 More linear scalability beyond 8-way, and especially beyond

16-way

 Server Consolidation

 Workloads leveraging large SQL Buffer Cache

 OLTP with large working sets

 Relational Data Warehouses

 64-bit eliminates performance overhead of AWE, the 32-bit

large-memory technology

SQL Server (64-bit)

64-bit address space

 Flat address space

 Applies to all memory

 DB Page Buffers need not

undergo AWE map/unmap

 Scarce resource on 32-bit is Procedure Cache

 Migrating to 64-bit and eliminating the bottleneck reduces

CPU and enables greater throughput

Consolidation / Proc Cache

40 60 80 100 120

TPM versus think time

CPU% and SQL compilations/sec versus think time

Example: Large-Scale OLTP Migration

 Large US Escrow / Title company

 Document management, title and tax processing

 High-volume, complex transactions and searches

 Refinance boom had stressed their 32-way 32-bit servers to over

70% CPU at peak

 Migrated OLTP from 32-way 32-bit to 32-way 64-bit

 Reduced CPU from 70% to 30%

 Indexing time improved by 30%

Example: Scale Up

Bulk Insert Throughput

16-way Unisys Cassin (Itanium II – 1Ghz) and16-way Unisys

ES7000 (1.6Ghz Xeon) both with 64GB of RAM

32-bit vs 64-bit BULK INSERT Throughput

10 20 30 40 50 60 70 80

Example: Eliminate AWE

Table Tot Rows Data Size(MB) Index Size (MB)

64 bit 64-bit Faster by 32 bit

16-way 36,120 2,994 2.07 1.87 74,905 5,601 4-way 30,470 7,931 1.85 1.99 56,313 15,819

Re-select data which is already memory-resident

Query: select sum(Qty) from ORDERS

16-way Unisys Cassin (Itanium II – 1Ghz) and 16-way Unisys ES7000 (1.6Ghz Xeon) both with 64GB of RAM (16GB to SQL)

Advantages Of Analysis Services (64-Bit)

 Not constrained by 3GB memory

limit

 Dimension memory + Shadow

dimensions + Buffer space (partitions)

 Cache fills

remaining space

 Huge dimensions supported in

memory

 MOLAP dimensions are extremely

large, but provide best query performance

Dimension Memory

Shadow Dimensions

Available Cache Processing Buffers

Advantages Of Analysis Services (64-Bit)

 More parallelism in partition processing is possible due to memory

advantages

 Each partition needs “working” memory

 Analysis Services can make use of huge memory available for

query cache and cube processing

 Faster aggregation processing

 Aggregations are built in memory during

cube processing, avoiding Temp Files

 Very large filesystem cache speeds cube query operations

Example: Analysis Services

OLAP Large Customer Dimension

 Transaction tracking to customer level – 3 Million key

Good SQL Server 64-Bit Candidates

throughput

Customer Examples / Deployed

 Leading Express Package Shipper

 Customer-level OLAP for profitability analysis / route cost

structures

 1+ TB source data

 Clalit Health Services

 Largest Healthcare provider in Israel

 Combined Relational and OLAP

 1+ TB

Customer Examples / Under

Development

 Large brokerage network

 Massive customer-level OLAP solution for assets and trading

activity

 Millions of products and Millions of accounts

 Property & Casualty Insurance provider

 Multi-TB relational analytics workload

 Initial workload benchmarks show 2.5x query improvement of

1.5Ghz Itanium II vs 1Ghz Xeon

 Beware the oversell:

 64 Bit ≠ 2x performance of 32 bit

 4-way Itanium II ≠ 8-way Xeon

64-bit Caveats

 Extended stored procedures and AS UDFs must be

recompiled for 64-bit

 ODBC Drivers for 3rd party databases

 Device drivers for SAN

 Management and monitoring tools