Transcript Grid Computing Meets the Database

Grid Computing Meets the Database
Chris Smith
Platform Computing
Session # 36686
The best thing about the Grid is that it is
unstoppable.
The Economist, June 21, 2001
2
© Platform Computing Inc. 2003
What is Grid computing?
Grid: Transparent, secure
and coordinated computing
resource sharing across
geographically disparate
sites
3
© Platform Computing Inc. 2003
Benefits of Grid Computing
Grid technology is used to aggregate computing resources across the
entire organization, regardless of location or business unit.

Provides virtually unlimited computing capacity

Delivers reliable, “always-on” computing infrastructure

Virtualizes IT infrastructure for end-users
Coordinates the usage of heterogeneous computing resources in
order to accomplish business processing tasks

4
© Platform Computing Inc. 2003
Example Use Cases
5

Batch Process Automation

Multi-Site Capacity Computing

Service Virtualization
© Platform Computing Inc. 2003
Batch Process Automation
What is Platform JobScheduler?
Intelligent batch process automation
Grid-enabled enterprise batch process automation software
Provides a Graphical Design Studio & Management console to
design and control the scheduling of Oracle jobs and compute jobs
with various dependencies (Line-of-Business Processes) across a
virtualized environment
7
© Platform Computing Inc. 2003
Simplified Scheduling Environment for
Oracle jobs and Compute jobs
Single Point of Control to Design & Monitor
Job Events,
File Events,
Time Events
Central Repository for Storing/Sharing
Jobs
Business flows
Sub flows
Proxy dependencies
Consistent, Flexible & Extensible Automated Exception Handling
Re-running jobs,
Killing jobs,
Triggering other jobs
8
© Platform Computing Inc. 2003
More Efficient Use of Computing Resources for
Oracle jobs and Compute jobs
Resource Virtualization
Ensures the reliability of mission critical business flows and alwayson availability of resources
Provision additional databases for specific tasks across time
Matching demand for resources with the supply of resources
9
© Platform Computing Inc. 2003
JobScheduler Architecture
Process
Designing/
Control
Load
XML
Save
XML
Client
10
© Platform Computing Inc. 2003
Scheduling
Time, Job, file,
Other events
Grid-Enabled Application
Execution
Infrastructure
Oracle
Database
Log
Jobflow Server
Grid Master & Grid Agents
JobScheduler and Oracle scheduler integration
2
3
Platform JobScheduler
server
LSF Master host
1
Platform JobScheduler client
elim.oracle.C
elim.oracle.B
orajobstart
4
LSF
Cluster
LSF host
Oracle client
C
11
© Platform Computing Inc. 2003
Oracle
instance
B
Oracle
instance
ETL using Platform JobScheduler
A common use of the Platform JobScheduler and Oracle scheduler
integration is for ETL into a data warehouse.
Example: a brokerage firm wants to load the day’s trading data into their
data warehouse for analysis (e.g. risk positions, trending, etc)

ETL flow is triggered by:
Time of day event
Arrival of market data in flat-file format
Completion of a stored procedure which collects location brokerage
data

Data is cleansed and loaded with SQL*Loader into the database
Stored procedures are invoked which do some analysis and initial
reporting

12
© Platform Computing Inc. 2003
Multi-Site Capacity Computing
Increasing Computing Capacity with Platform MultiCluster
A parameter space study is done on tens of thousands of individual sets
of parameters, resulting in tens of thousands of analysis jobs
Local cluster doesn’t have enough capacity, so Platform MultiCluster is
used to allow the forwarding of analysis jobs to clusters located at other
sites of the organization
The DBMS_STREAMS_ADM.MAINTAIN_TABLESPACES procedure
provided with Oracle Database 10g is used to replicate input data for the
analysis at the remote site
Database aware scheduling is used to make intelligent decisions about
which sites are suitable for receiving jobs
15
© Platform Computing Inc. 2003
Platform MultiCluster Job Forwarding Model
Send
queue
Compute
Servers
Site A
16
© Platform Computing Inc. 2003
Receive
queue
Compute
Servers
You submit
We do --• Job transfer
• data staging
• Account
mapping
• Accounting
Site B
Enterprise Grid Architecture
17
© Platform Computing Inc. 2003
Workload driven data management
1. Job forwarded
2. Run pre-exec
Master molecular
database (MOL)
3. Connect to MOL and run
MAINTAIN_TABLESPACES
Pre-exec
script
5. pre-exec finished
6. Job is run
Streams maintained
version of MOL
Streams DML updates
Application
Tablespaces
for MOL
7. Job uses copy
4. MOL metadata and
tablespaces transferred
Tablespaces
for MOL
18
© Platform Computing Inc. 2003
Database aware scheduling
Site 1 – MOL, MOL2
2. Update cache info
Site 2 – (none)
Data Management
Service
Site 3 - MOL
4. Local site is overloaded
Database aware scheduler
plug-in decides to forward
the job to site 3, since it
has the MOL database
Site 1
1. Poll for datasets
Site 3
5. Job forwarded to site 3
3. bsub -extsched MOL
MOL
MOL2
MOL
Site 2
19
© Platform Computing Inc. 2003
Service Virtualization
Demo Lab Hardware
-- A Common Web Service/Application Environment
Interconnect network
Storage network
Public network
Web
CISCO Hardware Load Balancer
Web Server &
App Server
Node
Node
Node
Node
Node
Node
Node
Node
(Linux)
Oracle RAC
(Linux AS 2.1)
NAS/SAN
21
© Platform Computing Inc. 2003
Oracle RAC Provisioning Demo System
Web Layer/Nodes
(Linux)
Node5
Agent
Manager
Node6
Node8
Agent
Manager
Service
Agent
Service
Agent
Web
Server
Apps
Apps
instances
App Server
Apps
Apps
instances
Managed node
Application Layer/Nodes
(Linux)
Provisioner
…
Agent
Manager
App
Agent
RAC
Agent
Applicatio
Apps
Apps
n instances
Managed node
Managed node
RAC
Apps
Apps
instances
RAC Layer/Nodes
(Linux AS 2.1)
22
© Platform Computing Inc. 2003
Managed
cluster
RAC
Node1
…
Node4
Proof of Concept Demos
Dynamic Provisioning within Database Layer
Dynamic Provisioning cross Database and Application Layers
23
© Platform Computing Inc. 2003
Provisioning Within DB Layer
Web Layer
Web Server
Node
-
App Layer
App Server
App Server
Node
Node
Node
RAC Layer
?
Show one RAC node running dbFinance, two RAC
nodes running dbHR, and one RAC node is idle
-
Have a lot of data access to dbFinance, a few
of data access to dbHR
-
Without dynamic provisioning, the response time to
dbFinance is very slow, while other RAC nodes are idle
-
Applying dynamic provisioning, one idle node is added to
dbFinance, one dbHR node is shutdown and moved to
dbFinance
-
The response time to dbFinance is improved
dbFinance
dbHR
Provisioning Across DB & App Layers
Web Layer
Web Server
Node
App Layer
?
Show one RAC node running dbFinance, one RAC
node running dbHR, and two RAC nodes are idle
-
Have a lot of applications need to run on App Layer
-
Without dynamic provisioning, the response time of
App Layer is very slow, while some RAC nodes are idle
Applying dynamic provisioning, some applications are
running on two idle RAC nodes
The response time of App Layer is improved
-
dbFinance
App Server
App Server
Node
Node
Node
-
-
RAC Layer
-
When there are some data accesses to dbFinance, more
database instances are needed
-
Applications on the RAC nodes are gracefully preempted,
and two more dbFinance instances are started
dbHR
App Server
App Server
RAC Agent
Gathers Metrics:
numInstances – Instances in a given database.
instanceState – Operation state of an instance.
dbLoad – Various load metrics from a database
User Calls, Recursive Calls
Physical Reads, Physical Writes
Consistent Gets, dB Block Gets
Takes Actions:
startInstance – Start an instance on a candidate
stopInstance – Stop an instance on a candidate
26
© Platform Computing Inc. 2003
Policy Functions
Discover State of System
What is the current state of the Candidates
Database High Load
If a candidate is free start an Instance of the loaded database.
Database Low Load
If a candidate was added, shutdown the database instance on
the candidate.
27
© Platform Computing Inc. 2003
Scenario 1: Results
Discovery
• Discover pe02, and pe03 are free
High Load
• Detect High Load on HR database.
• Have a candidate free.
• Remove candidate from free host list.
• Start another instance of the HR database.
• Add the candidate to the list of HR instances.
28
© Platform Computing Inc. 2003
Scenario 1: Results
Continued High Load
• Add the remaining candidate to the HR instances.
Low Load
• Detect low load on the HR database.
• Detect that candidate hosts are in use.
• Remove from last added candidate from list of HR instances.
• Stop HR instance on candidate.
• Return candidate to list of free hosts.
29
© Platform Computing Inc. 2003
Questions?