Background copy bandwidth impact on foreground I/O latency

The background copy bandwidth determines the rate at which the background copy for Active-active (HyperSwap®), Metro Mirror or Global Mirror Copy Services are attempted.

The background copy bandwidth can affect foreground I/O latency in one of three ways:
  • If the background copy bandwidth is set too high for the intersystem link capacity, the following results can occur:
    • The intersystem link is not able to process the background copy I/Os fast enough, and the I/Os can back up (accumulate).
    • For Metro Mirror and HyperSwap®, there is a delay in the synchronous secondary write operations of foreground I/Os.
    • For Global Mirror, the work is backlogged, which delays the processing of write operations and causes the relationship to stop. For Global Mirror in multiple-cycling mode, a backlog in the intersystem link can congest the local fabric and cause delays to data transfers.
    • The foreground I/O latency increases as detected by applications.
  • If the background copy bandwidth is set too high for the storage at the primary site, background copy read I/Os overload the primary storage and delay foreground I/Os.
  • If the background copy bandwidth is set too high for the storage at the secondary site, background copy write operations at the secondary overload the secondary storage. As result, the synchronous secondary write operations of foreground I/Os are again delayed.
    • For Global Mirror without cycling mode, the work is backlogged and again the relationship is stopped

To set the background copy bandwidth optimally, you must consider all three resources (primary storage, intersystem link bandwidth, and secondary storage). Provision the most restrictive of these three resources between the background copy bandwidth and the peak foreground I/O workload. You must also consider concurrent host I/O. If other write operations arrive at the primary system for copy to the remote site, these write operations can be delayed by a high level of background copy. As a result, the hosts at the primary site receive poor write-operation response times.

The provisioning for optimal bandwidth for the background copy can also be calculated by determining how much background copy can be allowed before the performance of host I/O becomes unacceptable. The background copy bandwidth can be decreased slightly to accommodate peaks in workload and provide a safety margin for host I/O.

Example

Consider the following example. The bandwidth setting at the primary site for the secondary clustered system is set to 200 MBps (megabytes per second) and the relationships are not synchronized. The system attempts to resynchronize the relationships at a maximum rate of 200 MBps with a 25 MBps restriction for each individual relationship. The following items can restrict throughput:
  • The read response time of back-end storage at the primary system
  • The write response time of the back-end storage at the secondary site
  • Intersystem link latency