April 28, 2015 – Advisory: Mirroring Default QoS Value Considerations
InterSystems has determined that the default value of the mirroring Quality of Service Timeout (QoS) setting is too small for some environments. This can result in undesired mirror failovers and/or unnecessary alerts.
This advisory applies to all platforms. Although the issue exists with any version of Caché, Ensemble or HealthShare that supports mirroring, this advisory is of particular note to deployments using version 2015.1 with the arbiter.
The Quality of Service Timeout (QoS timeout) setting affects failover member and arbiter behavior by defining the range of time, in milliseconds, that a mirror member waits for a response from another mirror member before taking action. The QoS timeout itself sets the maximum waiting time, while the minimum is one-half of that. A larger QoS timeout allows the mirror to tolerate a longer period of unresponsiveness from the network or a host without treating it as an outage; decreasing the QoS allows the mirror to respond to outages more quickly.
InterSystems has observed that virtualized hosts, in particular, often become entirely unresponsive for periods that exceed the default QoS value when operations such as backup or guest migration occur. Typically, deployments on physical (non-virtualized) hosts with a dedicated local network can operate within the default QoS value continually.
The default QoS timeout is 2 seconds (2000 ms). In future versions, the default QoS timeout will be 8 seconds (8000 ms) to allow for several seconds of intermittent unresponsiveness that may occur on some hardware configurations.
InterSystems recommends reviewing your hardware configuration and the effects of any operations regularly performed on the virtualization hosts to determine a reasonable setting for QoS. In deciding whether to increase the QoS timeout, the potential of an undesired failover should be weighed against a slower response time during actual outages. The timeout can be changed via the System Management Portal ‘Edit Mirror’ page, or via the ^MIRROR utility. If you have any questions regarding this advisory, please contact InterSystems’ Worldwide Response Center.