Changing the number of parallel threads for FLI
It might be needed to change the number of parallel threads for the Fast Link Interface (FLI).
FLI is used between instances of a cluster to exchange configuration data and transaction data. It keeps all instances of a cluster at the same configuration and data level.
Do not change the model revision or system configuration while you complete the steps.
Choose one of the following methods, depending on how much downtime is available.
Method 1
Using method one might cause some downtime, depending on the
setting on .- On , change to the new value.
- Restart one instance. After the instance restarts, its status is set to Synchronizing.
- Shut down all the other instances. Wait for the receive timeout for Status Control Interface (SCI) to expire. The status of the restarted instance changes to OK, and it can take transactions over Message Command Interface (MCI) and Batch Data Interface (BDI).
- Restart all instances.
Method 2
Using method two minimizes downtime.
- In Synchronisation threshold to a much higher value, for example, 10000000 messages. , change
- In , change to the new value.
- Restart one instance. Wait until it is restarted and the status is OK.
- Restart all other instances in the cluster one by one. Do not the delay the restart of each instance.
- Wait until all instances are synchronized. Change Synchronization threshold back to its original value.
It is a tuning process to find the optimal value for this setting in a specific environment. Running too many threads in parallel for Fast Link Interface might not always optimize the FLI synchronization:
- It might cause potential merging computation problems if the source and target transactions are processed in parallel by the Fast Link Interface. The setting resolves this issue.
- It might reduce the synchronization speed due to internal synchronization overhead.
Also, updates to system configuration, model revision, or investigation case first stop all transaction data synchronization, and are then performed single-threaded in FLI synchronization. If large configuration messages must be synchronized, increasing the number of threads for FLI does not improve the message processing rate.
To optimize the FLI throughput between two data centers in long distance, you can use the following formula to estimate the total number of FLI threads to set in the cluster:
<average FLI latency> = (<network latency> + <MCI average latency>) * 1.3
Expected FLI tps = (1000 / <average FLI latency>) * <FLI number of threads>
- The
avg rtt
from ping result can be used as the network latency. - MCI average latency can be captured by using IBM® Safer Payments KPIs.