Manage workloads across replicas via GDS Global Service attributes

Oracle Global Data Services (GDS) is a feature of Oracle Database 12c that provides connect-time and run-time load balancing, region affinity, replication lag tolerance based workload routing, and enables inter-database service failover over a set of replicated databases. GDS maximizes performance and availability of the applications and it allows database workloads to be managed across distributed replicas with a single unified framework.

Achieving the capabilities of GDS is as simple as setting the attributes of the global Services. There isn’t any custom code that needs to be written to attain these capabilities. The blog post lists some of the key attributes of global Services, which are set as part of the definition of the global Services.

Note:  All the attributes of the Global Service are applicable to both Oracle GoldenGate and Oracle Active Data Guard, except the Role and the Lag attributes – which are currently supported only for Active Data Guard.

Preferred, Available, Preferred_All: When a Global Service is created, we can define which databases support that service – known as the preferred databases for that Service. We can also define available databases to support a Global Service if the Service’s preferred database fails. If a service needs to be run on all the replicas of a given GDS pool, we can set the preferred_all attribute. Following are some examples for setting the preferred, available and preferred_all attributes for global Services:

GDSCTL>add service -service floor_report1 -gdspool sales -preferred sfo -available chi

GDSCTL>add service -gdspool sales -service sales_reporting_srvc  -preferred_all

If a database running a global service crashes, GDS taking into account the service placement attributes, automatically performs an inter-database service failover to another available database in the pool. It then sends Fast Application Notifications so that the client connection pools can reconnect to new database where the global Service has been started.

CLBGOAL: With the CLBGOAL attribute of a global Service, we can attain connect-time load balancing i.e. choosing the least loaded database instance for establishing a new connection. This is done for any client that uses a GSM to connect to a global service. GDS takes into account the load statistics from all GDS pool databases, Inter-region network latency, locality and CLBGOAL attributes.

CLBGOAL supports two values – LONG and SHORT.

Use the ‪LONG connection load balancing method for applications that have long-lived connections. This is typical for connection pools and SQL*Forms sessions. ‪LONG is the default connection load balancing goal.

Use the ‪SHORT connection load balancing method for applications that have short-lived connections. When using connection pools that are integrated with FAN, set the ‪CLBGOAL to ‪SHORT.

Here is an example for setting the CLBGOAL:

GDSCTL>add service -service sales_clb_srvc -gdspool sales –preferred_all -clbgoal LONG

RLBGOAL: Run-time load balancing is a feature of Oracle connection pools that can distribute client work requests across persistent connections that span databases. GDS supports Runtime load balancing feature of connection pool based clients (OCI, JDBC, ODP.NET, and WLS) that are integrated with the Oracle database, i.e. for a particular work request, “picking” a connection in the pool that belongs to the least loading instance. RLB does not require establishing of new connections and can be done much faster and more often.

GDS, as part of advisory compilation, takes into account per-service performance data from pool databases, inter-region network latency, locality and RLBGOAL settings. GDS compiles an array, such that each entry in the array corresponds to the percentage of work requests that ought to be sent to a particular database instance in the GDS Pool. GDS publishes this array every 30 Seconds to regional client connection pools as RLB event via ONS Channels. Based on the RLB advisory, clients distribute workload requests across persistent connections spanning GDS Pool database instances

RLBGOAL supports two values – SERVICE_TIME and THROUGHPUT. With the RLBGOAL set to SERVICE_TIME, Connection Pools route work requests to DB instances to minimize response time. The load balancing advisory is based on elapsed time for work requests using the service and network latency. With the RLBGOAL set to THROUGHPUT, Connection Pool routes work requests to DB instances to maximize total throughput of the system. The load balancing advisory is based on rate of work completion in the service plus available bandwidth to service. Here is an example for setting the RLBGOAL:

GDSCTL>add service -service sales_rlb_srvc -gdspool sales –preferred_all –rlbgoal SERVICE_TIME

However, when there is long term change in instance performance, there may be a need to redistribute pooled connections among the instances. The process of automatically moving unused connections from under-loaded to heavily loaded instances is termed as Gravitation. Sometimes when there are long-term changes in the database instance load, the client cannot follow RLB advisory because there are not enough connections to push the workload to a particular instance. At the same time another instance may have unused connections. The gravitation algorithm implemented in the connection pool detects the imbalance in the connection distribution, and creates or terminates connections from the database accordingly. GDS supports the gravitation feature of Oracle connection pools.

Besides supporting Gravitation, GDS proactively handles the instance UP and DOWN events sent through the ONS. It allows to quickly removing dead connections from the pool preventing the application from using them and also proactively establishing connections to an instance that was restarted or added to the pool.

When a GDS pool contains replicas hosted on database servers of different processor generations and different resources (CPU, memory, I/O), GDS understands how the Global services are performing across these replicas running on hardware with varying power.

In such scenarios, work requests are sent to less powerful databases, but not until more powerful databases are overloaded and aims to equalize the response time.

 

gds lb

Figure 1: GDS does intelligent load balancing even across asymmetrical database servers

Figure 1 showcases a benchmark against three database servers – ORLBb, ORLBc, ORLBd that have 4 CPUs, 3 CPUSs and 2 CPUs respectively. The top graph depicts the run-time load balancing information that is sent to connection pool based clients. Accordingly, higher percentage of workload is advised to be sent to ORLBb. The bottom graph demonstrates how GDS directed the client load in such a way that equalized the response time.

GDS understood that the faster hardware is able to crunch more workload and thus sends Runtime Load balancing information that advises the clients to send more work requests inclined towards the faster hardware relative to the slower resources.

This is a neat capability wherein customers can transparently phase in faster and efficient infrastructure resources and GDS allows them to take advantage of the newer hardware by pushing more workload to it and reaps the benefits of the efficient resources. Thus, GDS does intelligent load balancing even across asymmetrical database servers.

Locality: Customers can also set preferences on which region the given applications should connect to. With the Locality attribute, GDS controls geographical affinity between clients and databases.

With the Locality attribute set to ANYWHERE, the client connections and work requests are routed to the best database in any region that satisfies CLB and RLB goals for the given Global Service. For example:

GDSCTL>add service -service sales_reader_srvc -gdspool sales –locality ANYWHERE – preferred_all

With the Locality attribute set to LOCAL_ONLY, the client connections and work requests are routed to the best database in the client’s local region that satisfies CLB and RLB goals for the given Global Service. For example:

GDSCTL>add service -service sales_reader_srvc -gdspool sales –locality LOCAL_ONLY – preferred_all

With the Locality attribute set to LOCAL_ONLY -REGION_FAILOVER, the client connections and work requests are routed to the best database in the client’s local region that satisfies CLB and RLB goals for the given Global Service. If no DB in local region offers a service, client request is forwarded to the best database in another region. For example:

GDSCTL>add service -service sales_reader_srvc -gdspool sales –locality LOCAL_ONLY –region_failover –preferred_all

Lag (for Active Data Guard): In a replicated environment, the replica database may lag behind its primary database. With GDS, customers can choose the lag tolerance (in Seconds) that is acceptable for a given application. GDS routes requests to a standby whose replication lag is below the limit. When the replication lag exceeds the lag limit, the global Service is brought down and the new requests are routed to a database that satisfies the lag limit. In Oracle Database 12c Global Data Services, the lag attribute is supported for Active Data Guard only. An example for setting the lag attribute is shown below:

GDSCTL>add service -service sales_reader_lag180_srvc -gdspool sales -preferred_all -role PHYSICAL_STANDBY -lag 180

Role (For Active Data Guard): The role attribute is applicable to the GDS pools containing an Oracle Data Guard Broker configuration. When the role attribute is set, GDS makes sure that it can be started on a database if its role matches global service’s role attribute. GDS provides this functionality without requiring Oracle Clusterware to be installed.

GDSCTL>add service -gdspool sales -service sales_reporting_srvc  -preferred_all -role physical_standby

If the -role option is PHYSICAL_STANDBY, GDS allows the use of -failover_primary option to enable the service to failover to the primary database, if all standby databases are down. But, if there are other standby databases that are available, then those will be given the priority for the service to failover to before trying to failover to the primary database.

GDSCTL>add service -gdspool sales -service sales_reporting_srvc  -preferred_all -role physical_standby –failover_primary

For both manual and Fast-Start Failover (FSFO) based data guard role transitions, GSMs ensure that appropriate global Services are started based on the specified role. Fast Application Notification (FAN) events are published after a role change. The FAN events enable Fast Connection Failover (FCF) of client connections to an appropriate database instance within GDS Pool.

As I stated earlier, attaining GDS capabilities is as simple as setting the attributes of the global Services. There isn’t any custom code that needs to be written to attain these capabilities. For the complete list of all attributes of a global service, refer to the Oracle GDS documentation.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s