Netmail Store includes an adaptive power conservation feature that supplements Netmail Store's naturally green characteristics. This power-saving mode, also referred to as Darkive™, spins down disks and reduces CPU utilization after a configurable period of inactivity.
Note: The configuration options for power management apply to the entire cluster.
How Power Management Works
The power-saving mode causes a node with no incoming SCSP requests (both client and internode) in the last configurable
power.sleepAfter seconds to change to an Idle status in the Admin Console and pause its health processor. There is usually a delay between the node Idle status and its volumes Idle status as in-process replications are performed between nodes even after they are idle to ensure full data protection. After all queued activity is completed, the disks will eventually spin down (if supported by the disk manufacturer) and display as Idle if no further activity causes disk I/O.
When it appears in the Netmail Store Admin Console, Idle has different meanings for a node and volume:
- Idle node. A node with no SCSP activity for a specified length of time. In an idle state, a node's health processor pauses so an idle node is more likely to have idle volumes.
- Idle volume. A volume with no I/O activity for at least six minutes.
The cluster automatically awakes one or more nodes to carry out requests when needed and eventually revives all nodes if required. If there is no intervening activity after the configured
power.wakeAfter time period, the cluster will awaken all nodes to perform data, disk, and replication integrity checks.
Best Use of Power Management
A cluster that is constantly in use may not benefit significantly from the adaptive power feature. However, a cluster with long periods of inactivity on nights and weekends can expect significant power savings using this feature. Because only inactive nodes are affected, maximum available throughput is not affected as well, although additional latency is incurred on the first access to a node.
Tip: To maximize power savings, do no set the
power.wakeAfter value too close to the
power.sleepAfter value or the disks will cycle quickly between sleeping and waking.
Note: If your cluster is in constant use (24x7) or if uninterrupted feed restarts are critical for your operations, use Full Performance Mode.
Setting a Node to Archive Mode
node.archiveMode setting allows an administrator to designate a new or empty node as an archive node that remains idle in low-power mode without participating in cluster activity until its capacity is needed. This allows administrators to have additional storage capacity online and available without paying for the associated power costs of the additional nodes.
Once an archive node is activated by diminishing capacity in the remaining cluster, it attempts to aggressively fill itself to capacity with incoming write requests. Once its storage capacity is full, the node returns to an idle state until its stored content needs to be read.
This option determines how long a period of inactivity should occur with no incoming SCSP requests before Netmail Store pauses the node's health processor and displays it as Idle in the Admin Console. If you select Full Performance Mode on the Settings page in the Admin Console, a node never displays as Idle.
Setting this value to a long period prevents Netmail Store nodes from becoming idle and taking advantage of power savings. Setting the value to a small number allows nodes to become idle after a reasonable period of inactivity (two hours by default). The smallest supported value is 60 seconds.
This option determines how long a node remains idle before Netmail Store wakes up an idle node to allow the health processor to validate disk content integrity and replicas. Setting this value to a long period can unnecessarily put content at risk because the health processor does not run on sleeping nodes. Setting the number to a small number reduces power savings, although the volumes and nodes return to an Idle state if additional client activity is not received.
See [health] .startDelay for information about the health processor.