Workload Management in ASE Shared Disk Cluster A Live Demonstration
Dec 18, 2015
Workload Management in ASE Shared Disk ClusterA Live Demonstration
Agenda
• Workload Manager Value Proposition• Demo Topology• Demo
• Note that this presentation assumes some basic knowledge of ASE Cluster Edition
– Hopefully you’ve attended the overview session we’ve just had (or the ISUG Techcast, or last year’s presentation, etc…)
ASE Cluster Edition Workload Management
Application and deployment scenarios vary wildly.
Our approach: Customizable workload management that tailors ASE to its environment
Workload management allows you to see the cluster from the perspective that matters – the application perspective
• “Application” based routing of incoming connections
• Failover and availability policy specified at the application level
• Load balancing strategy that fits the application
• Planned downtime with applications in mind
Applications and Connection Distribution
The integrated workload manager provides an application management layer
• Applications are hosted on one or more physical instances– ..and each instance can host multiple applications
• New driver technology allows ASE to route connections in the cluster
– No client changes, single system image maintained– Routing based on a desired application– Applications spanning multiple nodes can use load distribution policies or a
round-robin scheme
Application Hosting
Sales DB
Shared Disk Storage
Sales Logical Cluster
Finance Logical Cluster
Manufacturing Logical Cluster
Manufacturing DB Finance DB
ManufacturingApp Client
FinanceApp Client
SalesApp Client
4 8 4 6engines
SalesApp Client
ase1 ase2 ase3 ase4
Application Level Failover Rules
Applications have differing business priority, and therefore differing failover requirements. Different sites want different availability models.
Our approach: Support failover rules at the application level.
• Failover rules allow for the implementation of nearly any availability model (1+1, N+1, N+M, etc.)
• Out of the box setup fails over an application to the least loaded system
• Alternatively, you can specify specific failover rules – Mission critical applications can use resources from less important apps, but not the other way
around
Application Failover Example
Possible Scenarios• ase2 fails, Manufacturing
runs only on ase3• ase2 and ase3 fail,
Manufacturing runs on ase4• ase2 Fails, Manufacturing
runs on ase3 and ase4• ase4 fails, Finance runs on
ase1• ase1 and ase4 fail, Sales
runs on either ase2 or ase3, but Finance goes offline.
Sales DB
Shared Disk Storage
Connection Director
Sales Logical Cluster
Finance Logical Cluster
Manufacturing Logical Cluster
Manufacturing DB Finance DB
ManufacturingApp Client
FinanceApp Client
SalesApp Client
4 8 4 6engines
SalesApp Client
ase1 ase2 ase3 ase4
Load Distribution in Cluster Edition
Applications have different transaction profiles. A one-size-fits-all approach to load balancing doesn’t work.
Our approach: Support application level load distribution policies
• Load profiles describe what performance metrics matter to an application and how ASE should react to them
• OLTP and DSS optimized profiles are included, others can be created
• Load scores provide fast, simple mechanism for load monitoring and trending
Handling Planned Downtime
Most downtime is planned.
Our approach: Provide a graceful way to move applications
• Dynamic relationship between applications and instances hosted time• Administrator tells ASE about the need for downtime and gives a
deadline– ASE rehosts the application on another cluster instance
– New connections are routed to new instance
– Existing connections have until the deadline to commit transactions
– As xacts commit, connections are gracefully migrated to the new instance
Agenda
• Workload Manager Value Proposition• Demo Topology• Demo
aselakes Cluster
• Four instance cluster– aselakes1 on node como– aselakes2 on node harriet– aselakes3 on node phalen– aselakes4 on node crosby
• Three application DBs– Orders, Catalog, and
Development
Order DB Catalog DB Development DB
aselakes1 aselakes2 aselakes3 aselakes4
como harriet phalen crosby
Logical Cluster Configuration
• Order Logical Cluster– OLTP– runs on aselakes1
• Catalog Logical Cluster– Read-only app– runs on aselakes2 and
aselakes3
• Development Logical Cluster
– used for everything else, runs on aselakes4
Order DB
OrderLogical Cluster
Development Logical Cluster
Catalog Logical Cluster
Catalog DB Development DB
Catalog Client Dev Client
Order ClientOrder Client
aselakes1 aselakes2 aselakes3 aselakes4
Agenda
• Workload Manager Value Proposition• Demo Topology• Demo