Andrew Mendelsohn
高级副总裁
数据库
甲骨文公司
Oracle 数据库 10g：
数据库技术的革命
Oracle 数据库 10g 的目标
最高的服务质量
最高的可用性、可靠性、安全性
最高的性能、可伸缩性
使成本减半
降低硬件、软件、人力成本
最易于管理
Cut 硬件 cost by …
Cut 软件 cost by …
Cut labor cost by one-half to one-tenth
“今天的 IT 企业效率低下到令人无法忍受的程度，它们仅使用了其容量的一小部分。”
Frank E. Gillett
Forrester Research，2002 年 10 月
Traditional IT infrastructures are not making the best use of available computing resources. This means that globally, 10’s of billions of dollars are being wasted every year!
Enterprise 网格计算 with Oracle 10g eliminates that wastage by delivering enterprise applications 按需提供容量 through pooling and sharing of computing resources.
[next slide]
Enterprise
A business organization, in either the public or private sector
Enterprise Computing
Computing to support the operation of the enterprise as a public sector or private sector business
问题：计算孤岛
昂贵、未充分利用的组件
有限的可靠性
分散的管理
为适应业务需求而进行的调整非常缓慢
财务管理
采购
服务
供应链
文件
销售
HR
Cost - High
Biggest Computer = Most 昂贵 Computer
Capacity - Limited
Applications Outgrowing the Biggest Server
10,000 Global ERP users, 40,000 Email users
$40,000 per Processor
可靠性 - Limited
One Big Computer = Single Point of Failure
Fragmented Management
separate mgmt infrastructures (no 网格控制 to manage across)
Fragmented 安全性 (no single signon, scalable id mgmt)
Slow to adjust
Escalating 成本 – dedicated 硬件, 昂贵 组件, configured for peak load, high incremental 成本 due to limited 可伸缩性
The ad hoc infrastructure built up by IT over the years has created islands of computation. Each enterprise application has its own dedicated server and storage. As application demands fluctuate over time, one server can be at maximum usage while another server is sitting idle. To combat this, customers purchase extra server and storage capacity to meet peak demand. The HW is sized for the peak load of a particular application, which usually means one applications HW may be minimally used while another applications HW is maxed out. In fact, Giga Research says that the industry average for server utilization is about 30%. So that mean a lot of wasted capacity. And it’s not just the 硬件. 软件 and labor cost more since they also must be configured for peak loads. 硬件 (21%), 软件 (24%) and system administrator/DBA labor (40%) consume 85% of the typical IT budget.
And there is no inherent 可伸缩性 here. That’s b/c with a large SMP box, when companies reach the capacity of their server, they need to purchase an entirely new larger SMP box to run their application.
And then from an 可用性 standpoint, running apps on an SMP box, means a single point of failure (as we can see from the supply chain example here), unless companies have a backup 硬件 in place. So, uptime or 可用性 is often far less than optimal is these distributed environments. 安全性 also suffers because every system must maintain its own secure infrastructure. The individual islands inevitably result in multiple passwords, inconsistent 安全性 policies and varying levels of 安全性 management expertise across the systems.
[Over last five years, the rapid changes in technology have made possible a new way to run enterprise data centers.
联网的存储器
Powerful, industry standard 2-4 way servers, and blade servers
Commodity lan technology via ethernet
Gig Ethernet as commodity cluster interconnect
Blade servers put many tech together to put cluster in a box
[Click to next slide…]
技术趋势
联网的存储器
高性能的行业标准服务器
Linux
商用硬件联网
商用集群互联 — Gig E（千兆以太网）
整体式集群 — 刀片式服务器
解决方案：网格计算
将计算组件整合到少量的大型资源池中
计算组件、配置和管理过程的标准化
动态供应计算组件，以满足应用程序的需求和业务优先级
Enterprise
A business organization, in either the public or private sector
Enterprise Computing
Computing to support the operation of the enterprise as a public sector or private sector business
网格计算:
The use of provision-able resources to dynamically meet the requirements of applications and the priorities of the business
The consolidation of computing 组件 into a smaller number of larger resource pools to enable easier provisioning
The standardization of computing 组件, configurations, processes and applications within an Enterprise
The creation of an architecture that scales as resources grow [there’s also ha, 安全性, ease of management, … just means high QoS]
网格计算消除了计算岛
昂贵的组件
针对高峰负载而配置
高增量成本
有限的可靠性
分散的管理
调整缓慢
低成本、模块化组件
按需提供容量
低增量成本
容错
统一的管理
标准的配置
网格解决方案
IT 问题
网格计算 is a new computing architecture that addresses this problem. 网格计算 is defined as the coordinated use of many small servers acting as one large computer to run standard enterprise applications. 网格计算 has many advantages:
While existing infrastructure has islands of servers, an enterprise grid groups resources together and shares those resources among applications.
Existing infrastructure have high cost 硬件 and 软件 due to complexity and inefficiency. A grid can be built using off-the-shelf, 低成本 组件 such as Intel blade servers, the Linux operating system, and 模块化的storage. the biggest reason why the 软件 cost goes down is that the 硬件 utilization (per CPU) goes up.
Existing infrastructure can be difficult to reconfigure as workloads, particularly peak loads, change. With a grid, resources can be allocated as needed for changing workloads.
Existing infrastructure is difficult to manage because each system is unique. With an enterprise grid, all systems are standardized and managed as one resource.
介绍
Oracle 网格计算
最高的服务质量
使成本减半
最易于管理
存储器网格
数据库网格
应用服务器网格
网格控制
Oracle has created a 网格计算 infrastructure 软件 that balances all types of workloads across servers, and enables all those servers to be managed as one complete system.
网格计算 can achieve the same very high level of 可靠性 as a mainframe because all 组件 are clustered. But unlike mainframes and large UNIX SMP servers, a grid can be built with open systems technologies such as Intel processors and the Linux operating system so it’s very 低成本.
Oracle 存储器网格
存储器网格
数据库网格
应用服务器网格
网格控制
Oracle has created a 网格计算 infrastructure 软件 that balances all types of workloads across servers, and enables all those servers to be managed as one complete system.
网格计算 can achieve the same very high level of 可靠性 as a mainframe because all 组件 are clustered. But unlike mainframes and large UNIX SMP servers, a grid can be built with open systems technologies such as Intel processors and the Linux operating system so it’s very 低成本.
自动存储管理 (ASM)
不再需要传统的文件系统和卷管理器
按需提供容量
在线添加/删除磁盘
自动 I/O 负载均衡
跨磁盘为数据创建条带，以平衡负载
最高的 I/O 吞吐量
自动镜像
易于使用
自动存储管理
Portable, 高性能 cluster file system
No volumes: just a pool of storage
Partitions total disk space into uniform sized megabyte units
Units assigned to a file are tracked using database oriented indexing techniques
Indexing allows efficient add/remove of disk to stripe set with automatic rebalancing
Data is spread across disks to balance load
Integrated mirroring across disks
自动存储管理 (ASM)
自动存储管理
不再需要传统的文件系统和卷管理器
按需提供容量
在线添加/删除磁盘
自动 I/O 负载均衡
跨磁盘为数据创建条带，以平衡负载
最高的 I/O 吞吐率
自动镜像
易于使用
Portable, 高性能 cluster file system
No volumes: just a pool of storage
Partitions total disk space into uniform sized megabyte units
Units assigned to a file are tracked using database oriented indexing techniques
Indexing allows efficient add/remove of disk to stripe set with automatic rebalancing
Data is spread across disks to balance load
Integrated mirroring across disks
自动存储管理小结
低成本
不再需要卷管理器和文件系统
使用廉价、模块化的存储器有效工作
更高的存储器利用率
易于使用 — 数据库管理员/系统管理员工作量减少 50%
容错
原始磁盘 (Raw disk) 性能
按需提供容量
自动 I/O 负载均衡
Better utilization via no fragmentation
存储器网格 — 以最低的成本实现最高的数据保护
数据卫士
站点故障保护
Flash Forward Recovery is a new disk based B&R strategy
Backup and Recovery goes from hours to minutes
Trades cheap disk space for 昂贵 people time
Oracle offers a continuum of data protection technologies
ASM Mirroring - protects against disk failures
Flashback database - protects against logical failures
Disk Based Backup - protects against corruptions
Data Guard - protects against site failures
They can be easily combined to provide complete data protection
They all work transparently on clusters
Combine the features to achieve any level of data protection
Oracle 数据库服务器网格
存储器网格
数据库网格
应用服务器网格
网格控制
Oracle has created a 网格计算 infrastructure 软件 that balances all types of workloads across servers, and enables all those servers to be managed as one complete system.
网格计算 can achieve the same very high level of 可靠性 as a mainframe because all 组件 are clustered. But unlike mainframes and large UNIX SMP servers, a grid can be built with open systems technologies such as Intel processors and the Linux operating system so it’s very 低成本.
数据库服务器网格 —
真正应用集群
运行真正的应用程序
经实践验证
成千上万的生产站点
经主流应用程序认证
在标准、低成本的服务器上运行
容错
HR
财务管理
DSS
以低成本实现最高的性能
来源：Transaction Processing Council，截至 2004 年 1 月 1 日：Oracle 数据库 10g 与真正应用集群，16 个节点的 4 路 HP Integrity rx5670，1,184,893 tpmC，$ 年 4 月 30 日提供。在 64 路 HP Integrity Superdome 上的 Oracle 数据库 10g，1,008,144 tpmC，$ 年 4 月 14 日提供。
$
$
16X4 路 Itanium 2
1X64 路 Itanium 2
56
6
5
24
25
数据库服务器网格 —
集成的集群件
可移植至所有平台
低成本
无需购买额外的软件
易于安装、管理
容易
在几小时而不是几天内完成部署
改善的单系统镜像
单供应商支持
对第三方集群件保持开放
Multiple tier awareness
数据库服务器网格 —
工作负载管理
以按键的方式在线地在集群中添加/删除服务器
自动负载均衡
基于“服务”理念
自动将服务连接请求路由至拥有最低负载的适当服务器
一旦出现服务器故障，自动把幸存的服务器重新分配给服务
网格容错
Oracle 10g 基础架构最快的端到端应用程序故障切换
数分钟
数秒
数分钟
数秒
Oracle9i
Oracle 10g
数据库 应用服务器 总恢复
故障切换 检测 时间
数秒
< 1 分钟
Oracle9i does have mid-tier server failure notification, but it is not integrated into the iAS jdbc connection pooling as in 10i. 9i java app servers detect server failure via a tcp/ip timeout which is typically around 10-15 minutes (user configureable).
中国的众多 RAC 客户
中国移动
中国联通
广发证券
Hutchison-Priceline
铁道部
上海市电力局
清华大学
…
铁道部
运输管理信息系统 (TMIS)
铁路货运业务
列车调度
位于中国大型火车站中的 62 个 RAC 数据库
RAC 数据库的特性
每个数据库多达 4 个节点
大部分为每个节点 4 个 CPU
50-100 GB in size. 高性能 & HA.
Used for shipment tracking, cost calculation, best transfers
No central site, but 10 regional sites
Why 62 systems:
Biggest railway stations (62 most important cities)
62 RAC sites using either IBM HACMP or HP TRU64.
RAC is their standard.
Most sites are 3 nodes. One site is 4 nodes.
Some of them were migrated from 8i OPS to 9iRAC.
Some of them were running single instance and run out of the capacity, so they added the additional nodes.
System used for both HA and 可伸缩性.
中国移动
应用程序包括
预付费移动呼叫授权
计费
客户关怀
财务管理
数据仓库
许多省（例如，江苏、浙江、湖北）已经实施了 RAC 数据库
RAC 数据库的特性
每个数据库多达 4 个节点
每个节点 8 到 32 个 cpu
每个数据库多达 1300 万用户
每个数据库的数据多达 21 TB
Jiangsu Mobile Communications & ZheJiang Mobile Communications are the two who have implemented, these are 2 of the 3 biggest provinces in China.
JiangSu Mobile – HP UX – 10 RAC Clusters
Reporting (16* TB initial, 3 – 24 way nodes) - 13 million users
3 x Prepay Mobile Call Authorization ( TB, 2 nodes, 4000 SQL/sec) – 13 million users
ZheJiang Mobile – HP-UX – 3 RAC Clusters
Prepay Mobile Call Authorization ( TB, 4 nodes, 4000 SQL/sec) – 13 million users
Batch Billing (21 TB in just data, 4 – 16 way nodes, 4000 SQL/sec). 300 GB loaded per day.
Hubei Mobile – AIX P690 – 2 RAC Clusters
Prepay Mobile Call Authorization ( TB, 2 nodes, 9000 SQL/sec) – 7 million users
Batch Billing (10 TB , 2 – 16 way nodes, 48 GB RAM, 600 SQL/sec)
Financials (15 TB), DW (10 TB)
HP 8-32 way Superdome, 9 clusters, 2-4 nodes
Oracle RAC 10g — 企业网格的 唯一解决方案
行业标准的低成本服务器上的关键任务的服务质量
集成的集群件使得 RAC 易于被所有用户使用
容错，扩展所有应用程序
按需提供容量
自动负载均衡
Add first-failure diagnostics
Oracle 网格控制
存储器网格
数据库网格
应用服务器网格
网格控制
Oracle has created a 网格计算 infrastructure 软件 that balances all types of workloads across servers, and enables all those servers to be managed as one complete system.
网格计算 can achieve the same very high level of 可靠性 as a mainframe because all 组件 are clustered. But unlike mainframes and large UNIX SMP servers, a grid can be built with open systems technologies such as Intel processors and the Linux operating system so it’s very 低成本.
Oracle 10g 网格管理
一半到十分之一的成本
自我管理的数据库
OEM 网格控制
网格范围的操作
利用少量的增量成本从一个系统扩展到成百上千的系统
网格控制 provides
Entire stack
User Provisioning
Cloning
Patch
Management
业务挑战 —
如何将 DBA 的工作扩展到网格上？
软件
维护 6%
持续的系统维护 55%
安装 6%
其他
15%
来源：IOUG 2001 数据库管理员调查
数据加载 6%
Oracle 数据库 10g
自我管理的数据库
内置的智能基础架构
代码工具
工作负载信息库
日常任务自动化
自动的基于磁盘的备份和恢复
自动的优化器统计信息收集
自动的 PGA 和 SGA 内存管理
自动存储管理
为 DBA 提供支持的工具
自动的数据库诊断监视器
自动的调整优化器
更多其他特性……
RBO RIP
Automatic memory tuning – sql execution memory in 9i, shared memory pools in 10g
“高级的简单化。”
Infoworld
来源：Infoworld，2004 年 3 月 22 日
38
2
1
比 SQL 服务器更易于管理
所需的
DBA 时间
少于
MS SQL
的百分比
- 0%
- 76%
来源：Edison Group，比较管理成本研究
- 10%
- 50%
Oracle OEM 10g 网格控制
管理 Oracle 网格环境
扩展至成千上万的服务器
配置管理
定义和实施标准配置
供应
克隆
补丁管理
应用程序服务水平管理
Key goal – reduce labor cost by factor of 2 to 10
Complete and integrated
Manage all oracle products and key dependent data center infra 组件
Administration, monitoring, provisioning
Scale to thousands of servers
Manage groups of systems as one
(admin?; monitor aggregated perf rollups, thresholds, alerts and then drill down to identify problem; provision) “monitor - all that matters is monitor app svc level and then have ability to diagnose and fix”
Automated and policy based operations (define and enforce standards)
安全性 – user privilege, process controls, 安全性 alerts and patches (lje – treat 安全性 as full separate topic from oem)
Policy violation reporting
Configuration
Oracle best practices (db), ., multiplex control file
Patch and clone is key
Provisioning – clone, patch, add/remove nodes/stg (asm); group op for all of this (clone, patch, deploy agent, secure)
Centralized and remote operations
“Oracle 企业管理器 10g 提供了单一集成的解决方案，以从一个 HTML 控制台管理成千上万的系统。”
Warwick Calkin
National Manager
Telstra IT Infrastructure and Operations
Old quote: “We now can manage hundreds of servers throughout the world from within one powerful management console.”
We will use Oracle's new management 软件 to maximize the 可用性, 性能, and 可靠性 of Telstra's entire environment
网格上的商务智能
更低的成本
利用硬件、软件、人力节省
管理海量数据
按需提供智能
跨平台的可移动表空间
异步更改数据捕获
从您的数据中获取更多的价值
成熟的 OLAP、数据挖掘
改善的 SQL 分析 — SQL 模型子句
Additional integrated BI capabilities – SQL Models, olap and data mining extensions
HTML DB —
快速的 Web 开发工具
构建数据库驱动的 Web 应用程序
面向非编程人员
应用程序开发人员和应用程序用户只需要一个 Web 浏览器
解决 MS Access 激增的问题
支持托管开发和部署环境
Accelerated web development: Every organization has a wish list of Web applications it wants to build to track information and share data. This presentation discusses how Oracle HTML DB helps you consolidate and secure data currently stored in spreadsheets and personal databases and provide Web access to this data through accelerated development of database-driven Web applications.
Oracle 数据库 10g 性能量度
PLSQL 快 2 倍
表扫描快 2 倍
大规模 SQL 配置文件调整
浮点数学运算快 10 倍
批量 LOB 更新快 3 倍
导入快 15 倍
导出快 35%
增量备份快 10 倍
并行介质恢复快 2-4 倍
备份压缩快 17 倍
无限带宽互联
½ 的延迟，快 2 倍的块传输
TPC-C
在 RAC 上为 120 万，100 万 SMP
绝对最好，在 4 路 Intel 上最好
在相同的硬件上提高 28%
Bulk LOB Update up to 3x faster
Up to 10x less shared memory usage for querying highy partitioned objects (. a table having thousands of partitions).
Improved 可伸缩性 and 性能 of operations on objects with very large number of partitions, due to better memory usage. For example, we have tests showing more than 55% improvement in elapsed time from 9i to 10g for dropping a table with 21,504 partitions (Lineitem table from TPC-H schema, using composite range-hash partitioning)
Partition-aware Materialized Views fast refresh: 性能 of PCT refresh is enhanced using functional dependencies and query rewrite. We can provide incremental refresh due to changes in the partitioning schemes in the underlying table where it was not possible with 9i. In some cases, we see 46% improvement in elapsed time compared to 9i.
Between 40% and 60% for single table scans with simple predicates or no predicates. For example, select count(*) from table
would improve by about 40% CPU-wise
轻松过渡到网格
Oracle 10g 能够保护投资
在集群上运行所有现有应用程序
将数据转移至新的平台，速度和 FTP 一样快
低风险 — 从小规模开始， 按需增加容量
ERP
电子邮件
DW
Oracle 10g 网格计算
最高的服务质量
使成本减半
低成本, 模块化的存储器和服务器
完整、集成的软件基础架构
使人力成本减半
快速地调整以适应业务需求
按需提供容量
负载均衡
易于管理
“网格的最大优势在于它是无法阻挡的”
The Economist
2001 年 6 月
2
Within 10 years, every company will use enterprise 网格计算. The best part is that you can get started on savings today. Standardize on 低成本 technology and cluster your servers. Begin consolidation. Automate your management framework. All these things can be done now with Dynamic Computing technology from Oracle.
A
Cut 硬件 cost by …
Cut 软件 cost by …
Cut labor cost by one-half to one-tenth
Traditional IT infrastructures are not making the best use of available computing resources. This means that globally, 10’s of billions of dollars are being wasted every year!
Enterprise 网格计算 with Oracle 10g eliminates that wastage by delivering enterprise applications 按需提供容量 through pooling and sharing of computing resources.
[next slide]
Enterprise
A business organization, in either the public or private sector
Enterprise Computing
Computing to support the operation of the enterprise as a public sector or private sector business
Cost - High
Biggest Computer = Most 昂贵 Computer
Capacity - Limited
Applications Outgrowing the Biggest Server
10,000 Global ERP users, 40,000 Email users
$40,000 per Processor
可靠性 - Limited
One Big Computer = Single Point of Failure
Fragmented Management
separate mgmt infrastructures (no 网格控制 to manage across)
Fragmented 安全性 (no single signon, scalable id mgmt)
Slow to adjust
Escalating 成本 – dedicated 硬件, 昂贵 组件, configured for peak load, high incremental 成本 due to limited 可伸缩性
The ad hoc infrastructure built up by IT over the years has created islands of computation. Each enterprise application has its own dedicated server and storage. As application demands fluctuate over time, one server can be at maximum usage while another server is sitting idle. To combat this, customers purchase extra server and storage capacity to meet peak demand. The HW is sized for the peak load of a particular application, which usually means one applications HW may be minimally used while another applications HW is maxed out. In fact, Giga Research says that the industry average for server utilization is about 30%. So that mean a lot of wasted capacity. And it’s not just the 硬件. 软件 and labor cost more since they also must be configured for peak loads. 硬件 (21%), 软件 (24%) and system administrator/DBA labor (40%) consume 85% of the typical IT budget.
And there is no inherent 可伸缩性 here. That’s b/c with a large SMP box, when companies reach the capacity of their server, they need to purchase an entirely new larger SMP box to run their application.
And then from an 可用性 standpoint, running apps on an SMP box, means a single point of failure (as we can see from the supply chain example here), unless companies have a backup 硬件 in place. So, uptime or 可用性 is often far less than optimal is these distributed environments. 安全性 also suffers because every system must maintain its own secure infrastructure. The individual islands inevitably result in multiple passwords, inconsistent 安全性 policies and varying levels of 安全性 management expertise across the systems.
[Over last five years, the rapid changes in technology have made possible a new way to run enterprise data centers.
联网的存储器
Powerful, industry standard 2-4 way servers, and blade servers
Commodity lan technology via ethernet
Gig Ethernet as commodity cluster interconnect
Blade servers put many tech together to put cluster in a box
[Click to next slide…]
Enterprise
A business organization, in either the public or private sector
Enterprise Computing
Computing to support the operation of the enterprise as a public sector or private sector business
网格计算:
The use of provision-able resources to dynamically meet the requirements of applications and the priorities of the business
The consolidation of computing 组件 into a smaller number of larger resource pools to enable easier provisioning
The standardization of computing 组件, configurations, processes and applications within an Enterprise
The creation of an architecture that scales as resources grow [there’s also ha, 安全性, ease of management, … just means high QoS]
网格计算 is a new computing architecture that addresses this problem. 网格计算 is defined as the coordinated use of many small servers acting as one large computer to run standard enterprise applications. 网格计算 has many advantages:
While existing infrastructure has islands of servers, an enterprise grid groups resources together and shares those resources among applications.
Existing infrastructure have high cost 硬件 and 软件 due to complexity and inefficiency. A grid can be built using off-the-shelf, 低成本 组件 such as Intel blade servers, the Linux operating system, and 模块化的storage. the biggest reason why the 软件 cost goes down is that the 硬件 utilization (per CPU) goes up.
Existing infrastructure can be difficult to reconfigure as workloads, particularly peak loads, change. With a grid, resources can be allocated as needed for changing workloads.
Existing infrastructure is difficult to manage because each system is unique. With an enterprise grid, all systems are standardized and managed as one resource.
Oracle has created a 网格计算 infrastructure 软件 that balances all types of workloads across servers, and enables all those servers to be managed as one complete system.
网格计算 can achieve the same very high level of 可靠性 as a mainframe because all 组件 are clustered. But unlike mainframes and large UNIX SMP servers, a grid can be built with open systems technologies such as Intel processors and the Linux operating system so it’s very 低成本.
Oracle has created a 网格计算 infrastructure 软件 that balances all types of workloads across servers, and enables all those servers to be managed as one complete system.
网格计算 can achieve the same very high level of 可靠性 as a mainframe because all 组件 are clustered. But unlike mainframes and large UNIX SMP servers, a grid can be built with open systems technologies such as Intel processors and the Linux operating system so it’s very 低成本.
Portable, 高性能 cluster file system
No volumes: just a pool of storage
Partitions total disk space into uniform sized megabyte units
Units assigned to a file are tracked using database oriented indexing techniques
Indexing allows efficient add/remove of disk to stripe set with automatic rebalancing
Data is spread across disks to balance load
Integrated mirroring across disks
Portable, 高性能 cluster file system
No volumes: just a pool of storage
Partitions total disk space into uniform sized megabyte units
Units assigned to a file are tracked using database oriented indexing techniques
Indexing allows efficient add/remove of disk to stripe set with automatic rebalancing
Data is spread across disks to balance load
Integrated mirroring across disks
Better utilization via no fragmentation
Flash Forward Recovery is a new disk based B&R strategy
Backup and Recovery goes from hours to minutes
Trades cheap disk space for 昂贵 people time
Oracle offers a continuum of data protection technologies
ASM Mirroring - protects against disk failures
Flashback database - protects against logical failures
Disk Based Backup - protects against corruptions
Data Guard - protects against site failures
They can be easily combined to provide complete data protection
They all work transparently on clusters
Combine the features to achieve any level of data protection
Oracle has created a 网格计算 infrastructure 软件 that balances all types of workloads across servers, and enables all those servers to be managed as one complete system.
网格计算 can achieve the same very high level of 可靠性 as a mainframe because all 组件 are clustered. But unlike mainframes and large UNIX SMP servers, a grid can be built with open systems technologies such as Intel processors and the Linux operating system so it’s very 低成本.
56
6
5
24
25
Multiple tier awareness
Oracle9i does have mid-tier server failure notification, but it is not integrated into the iAS jdbc connection pooling as in 10i. 9i java app servers detect server failure via a tcp/ip timeout which is typically around 10-15 minutes (user configureable).
50-100 GB in size. 高性能 & HA.
Used for shipment tracking, cost calculation, best transfers
No central site, but 10 regional sites
Why 62 systems:
Biggest railway stations (62 most important cities)
62 RAC sites using either IBM HACMP or HP TRU64.
RAC is their standard.
Most sites are 3 nodes. One site is 4 nodes.
Some of them were migrated from 8i OPS to 9iRAC.
Some of them were running single instance and run out of the capacity, so they added the additional nodes.
System used for both HA and 可伸缩性.
Jiangsu Mobile Communications & ZheJiang Mobile Communications are the two who have implemented, these are 2 of the 3 biggest provinces in China.
JiangSu Mobile – HP UX – 10 RAC Clusters
Reporting (16* TB initial, 3 – 24 way nodes) - 13 million users
3 x Prepay Mobile Call Authorization ( TB, 2 nodes, 4000 SQL/sec) – 13 million users
ZheJiang Mobile – HP-UX – 3 RAC Clusters
Prepay Mobile Call Authorization ( TB, 4 nodes, 4000 SQL/sec) – 13 million users
Batch Billing (21 TB in just data, 4 – 16 way nodes, 4000 SQL/sec). 300 GB loaded per day.
Hubei Mobile – AIX P690 – 2 RAC Clusters
Prepay Mobile Call Authorization ( TB, 2 nodes, 9000 SQL/sec) – 7 million users
Batch Billing (10 TB , 2 – 16 way nodes, 48 GB RAM, 600 SQL/sec)
Financials (15 TB), DW (10 TB)
HP 8-32 way Superdome, 9 clusters, 2-4 nodes
Add first-failure diagnostics
Oracle has created a 网格计算 infrastructure 软件 that balances all types of workloads across servers, and enables all those servers to be managed as one complete system.
网格计算 can achieve the same very high level of 可靠性 as a mainframe because all 组件 are clustered. But unlike mainframes and large UNIX SMP servers, a grid can be built with open systems technologies such as Intel processors and the Linux operating system so it’s very 低成本.
网格控制 provides
Entire stack
User Provisioning
Cloning
Patch
Management
RBO RIP
Automatic memory tuning – sql execution memory in 9i, shared memory pools in 10g
38
2
1
Key goal – reduce labor cost by factor of 2 to 10
Complete and integrated
Manage all oracle products and key dependent data center infra 组件
Administration, monitoring, provisioning
Scale to thousands of servers
Manage groups of systems as one
(admin?; monitor aggregated perf rollups, thresholds, alerts and then drill down to identify problem; provision) “monitor - all that matters is monitor app svc level and then have ability to diagnose and fix”
Automated and policy based operations (define and enforce standards)
安全性 – user privilege, process controls, 安全性 alerts and patches (lje – treat 安全性 as full separate topic from oem)
Policy violation reporting
Configuration
Oracle best practices (db), ., multiplex control file
Patch and clone is key
Provisioning – clone, patch, add/remove nodes/stg (asm); group op for all of this (clone, patch, deploy agent, secure)
Centralized and remote operations
Old quote: “We now can manage hundreds of servers throughout the world from within one powerful management console.”
We will use Oracle's new management 软件 to maximize the 可用性, 性能, and 可靠性 of Telstra's entire environment
Additional integrated BI capabilities – SQL Models, olap and data mining extensions
Accelerated web development: Every organization has a wish list of Web applications it wants to build to track information and share data. This presentation discusses how Oracle HTML DB helps you consolidate and secure data currently stored in spreadsheets and personal databases and provide Web access to this data through accelerated development of database-driven Web applications.
Bulk LOB Update up to 3x faster
Up to 10x less shared memory usage for querying highy partitioned objects (. a table having thousands of partitions).
Improved 可伸缩性 and 性能 of operations on objects with very large number of partitions, due to better memory usage. For example, we have tests showing more than 55% improvement in elapsed time from 9i to 10g for dropping a table with 21,504 partitions (Lineitem table from TPC-H schema, using composite range-hash partitioning)
Partition-aware Materialized Views fast refresh: 性能 of PCT refresh is enhanced using functional dependencies and query rewrite. We can provide incremental refresh due to changes in the partitioning schemes in the underlying table where it was not possible with 9i. In some cases, we see 46% improvement in elapsed time compared to 9i.
Between 40% and 60% for single table scans with simple predicates or no predicates. For example, select count(*) from table
would improve by about 40% CPU-wise
Within 10 years, every company will use enterprise 网格计算. The best part is that you can get started on savings today. Standardize on 低成本 technology and cluster your servers. Begin consolidation. Automate your management framework. All these things can be done now with Dynamic Computing technology from Oracle.
