Building a Dynamic Grid Infrastructure

Agenda

• Dynamic Grid Infrastructure
  • What is it?
  • How is it different?
  • Why do we want it?
• Examples & Enablers
• Challenges for the Real Time Enterprise
• An Incremental Approach

Unsustainable & incapable of responding effectively to the rapidly changing needs and priorities of business
Silo infrastructure – inflexible, inefficient, lots of redundancy because of budget process and poor capacity planning, heterogeneity = complexity and cost in terms of skills and manual reactive processes
Monolithic apps – static, expensive customisation and integration of increasing number of packaged solutions
Grid exploits the concept of virtualisation which abstracts underlying technologies. This gives rise to the 3 key capabilities
Resource pooling and sharing – i.e. all resources including networks, storage, DB & mid-tier servers, data, function & workflow
Dynamic resource provision – allocate on demand,
Automation – respond automatically to meet defined service levels
Technical benefits –
Apps are decoupled from from static sever deployments and can share resources
Infrastructure can be flexed automatically in response to changing/unpredictable workloads

MercadoLibre is the eBay of LatinAmerica, and they are eBay’s partner in Latin America.
The whole environment from search, to bid, to customer care runs on a single cluster.
Originally MercadoLibre ran on a mid size SMP. Because of their rapid growth they needed more capacity and replaced this w/ a 4 node – 4way HP Linux Itanium clusters
As their business volume has kept expanding, they have been adding capacity incrementally and now have a 15 node cluster.Current plans are to add Montecito class nodes to this cluster, maybe replacing a few of the old nodes

This slide is important to help dispell the common myth that it is a gigabit ethernet will not be enough for most environments.Have 3 boxes with faster cpu. 1.6 Gb instead of 1.3.

Client integration with Load Balancing Advisory. No programming required, works out of the box if you use the Oracle driver for JDBC and ODP.NET. Must be using connection pooling.

Before load balancing advisory, connection pools randomly selected idle connections or select connections in a round robin when requests came in. Now we have intelligent allocation of connections. With JDBC, if a node is slow, hung or not responding for a long time, we will gravitate connections to other instances.

FINE GRAINED, AUTOMATIC RESOURCE MANAGEMENT
The Database Resource Manager, introduced in Oracle8i, has been significantly enhanced in Oracle9i to allow for more granular control over resources and adds features such as automatic consumer group switching, maximum active sessions control, query execution time estimation and undo pool quotas for consumer groups. Administrators are able to specify the maximum number of concurrently active sessions per consumer group. Once this limit is reached, Database Resource Manager queues all subsequent requests and runs them only after existing active sessions complete.
The automatic consumer group switching feature of Oracle9i allows the administrator to specify a certain criteria which, if met, will cause the Database Resource Manager to automatically switch the consumer group of a long running session, for instance, from a consumer group set up for OLTP operations to one more suited for batch reporting. Administrators are also able to set a maximum estimated execution time for each consumer group. The Database Resource Manager then estimates the approximate query execution time for each operation before it begins and will either abort the operation if it exceeds the limit specified or switch the operation into a lower priority consumer group. With the Undo pool quota feature, administrators are now able to specify a maximum on the total amount of rollback data generated per consumer group. This prevents a rogue transaction from consuming excessive rollback space and thus impacting system operation.

Resource Manager today does not
Manage IO
Have Fixed Caps (will not throttle until a capacity shortage exists)
Work across multiple instances on a node

BCF has been using Oracle database products since version 5 and have now implemented a full Oracle Grid architecture. The “Grid” consists of two 18-node RAC clusters connected by dark fibre about 10 KM apart. Each cluster has 1 node dedicated to a Dataguard standby for the databases on the remote cluster along with 2 nodes dedicated to backup and recovery. The 8 most critical SKU databases run on the two clusters along with other less critical databases. Instances are brought up on nodes as workload or maintenance requires.
They have started with a server consolidation approach by combining many separate 8i single instance servers and 9i RAC clusters hosting over 60 databases onto these two clusters. The longer-term plan is to further consolidate the individual databases into fewer larger databases.
They also combined all of the separate storage into one large ASM disk group along with a Flash recovery diskgroup. All data is now stored in ASM and storage is added and rebalanced as new databases are added or grow.
The entire grid is managed from a single point using Oracle Enterprise Manager Grid Control.

BCF has been using Oracle since version 5
BCF has been an early adopter of Oracle since the Sequent Numa dynix ptx nodes in the early 90’s
Has used technology as a differentiator in the marketplace, have been an alpha shop for Oracle AP and have worked closely with Oracle for many years
Existing db’s on 9i RAC on Suse Linux – migrating to 10g because 10g provided more of the infrastructure that we wanted, storage reconfiguration in 9i without ASM was tedious
The business requires pretty close to 24X7 uptime because there are various reports and data loads that are processed after hours and the databases are used as a mixture of OLTP of warehousing, I.e. a hybrid for retail processing, I.e. customer intelligence and OLTP stuff

2,000 concurrent users (employees) growing to support 200k concurrent external self service users
HP UX 11.11 Superdomes, 12 CPUs each
Peoplesoft CRM 8.9
Peoplesoft Portal 8.9
RAC 9.2.0.8
Financials and HCM are not RAC enabled at the moment

All the information on this slide is relevant to selling Grid on hardware partitioned servers. IT also important in convincing this is a better option than virtual partitioning

Oracle Grid builds on and enhances the benefits of hardware partitioning

This solution doesn’t deliver any additional benefits on one partitioned server

A minimum of 2 servers are required and ideally 3 because
2 servers in active-active mode can only use 50% of resources in order to provide equivalent failover

© 2010, www.oracledatabase12g.com. 版权所有.文章允许转载,但必须以链接方式注明源地址,否则追究法律责任.

相关文章 | Related posts:

1. 11gR2 Grid Infrastructure Redundant Interconnect and ora.cluster_interconnect.haip
2. How to Proceed from Failed 11gR2 Grid Infrastructure (CRS) Installation [ID 942166.1]
3. How to Troubleshoot Grid Infrastructure Startup Issues [ID 1050908.1]
4. 11gR2 Grid Infrastructure Single Client Access Name (SCAN) Explained
5. 11g Multi-Column Correlation Stats and Dynamic Sampling
6. HP-UX: Dynamic (runtime) Loading of Shared Libraries
7. Book:Building a WordPress
8. Building a RAC Test Environment on VMWare For Free
9. DBMS_SQL Package versus Native Dynamic SQL
10. Discover How To Leverage The Power Of The Grid