Improving performance for your storage system

There are tasks that you can do to improve the performance of your storage system. Performance factors include throughput speed, error rates, and failover recovery.

1. Monitor your current performance. See Exchange visibility.
2. Identify situations that are not occurring satisfactorily.
3. Deploy the IBM Health Center tool that is provided with the IBM JDK. Enable the tool on each member in the cluster through JVM start up arguments to identify performance inhibitors and resolve them.
   The following items are monitored by the Health Center:

   • Aggregate number of failed storage activities since the storage server started
   • Aggregate number of different activities in the last minute, the last hour, the last day, and since the reset button was pushed in the JMX tool (for ad hoc measurements). Activities since a reboot are not measured, unless the reset button was never pushed.
   • Success of the last reap cycle, including how many expired blobs were deleted
   • Indication that the memento file prevented reaping
   • Uptime
   • isRetired status
   • PUT and GET counts
   • DELETE count
   • Remaining usable space available for storage
   • Byte transfer counts for PUT and GET operations
4. Determine what is causing the delay and correct the situation.
5. For transfers exceeding 2 GB, the client should use the header:

   Transfer-Encoding: chunked

6. For transfers with more than 1024 open connections at the same time, increase the value of:

   ulimit -n

   The default value is 1024, but you can increase this value if you are receiving connection errors. However, if you allow too many simultaneous connections, throughput may be decreased.
7. Modify the lifespan of the blobs by configuring new purge settings to reduce the quantity of data in the storage file system.
8. Prior to shutting down an HTTP server used for storage, issue a quiesce command. This command stops all new incoming connections, and provides a grace period before interrupting in-flight connections that were established prior to the quiesce command.
   Important: HTTP Servers are automatically based on memory events, so that the system does not run out of memory. If system memory reaches the configured minimum level, a user interface warning is displayed to enable you to take corrective actions such as adding an additional node.
9. Test your new configuration to see if performance has improved.