Cluster interconnect

from Wikipedia, the free encyclopedia

The cluster interconnect is used for management, data transmission and load distribution purposes in a computer cluster and is a component of the cluster manager . In the literature, the cluster interconnect is sometimes referred to as the cluster heartbeat , although this function of the “heartbeat” is just one of several.

use

The interconnect of a cluster is a connection between the cluster members via which all kinds of relevant data are exchanged. It is usually designed as a private network segment in order to guarantee the highest possible operational security or to be able to map other network architectures . Because this interconnection should - depending on the size of the cluster - be equipped with the lowest possible latency and high transmission rate so as not to create a bottleneck. Frequently used technologies are Gigabit Ethernet or the more expensive InfiniBand .

As mentioned at the beginning, the cluster interconnect serves several purposes:

Heartbeat (Cluster Heartbeat)

Under Heartbeat ( dt. "Heartbeat" ) understands computer technology a periodically exchanged between related systems signal, analogous to the feeling of the human pulse . This signal is used for mutual monitoring: If the heartbeat of a system does not occur, an event is triggered to which the others can react accordingly. The cluster interconnect is used here for pure signal transmission. If the opposite side of the availability monitoring is not implemented by commercial clusterware , software packages such as LinuxHA are also available to secure individual services.

Until recently, only a serial cable was used for a pure heartbeat interconnect .

Cluster control

Another essential task of the interconnect is the management of the cluster. For example, the targeted switching of a service from one node to the other must be coordinated and take place with the least possible downtime. Each clusterware uses product-specific standardized telegrams of different sizes.

Data transfer / data balancing

When using the cluster for load balancing on application or database servers , it is also important in certain situations to pass on user data via the cluster interconnect. Here, too, the clusterware transmits special telegrams with the corresponding useful content, and an API is offered for the application above .

Example: Oracle RAC 10g uses what is known as cache fusion technology. It is based on the fact that blocks that have already been read into the cache of a cluster node can be obtained more quickly via the interconnect than by reading them again from the mass storage device . The sovereignty control for database blocks of this database management system also relies on the transmission of the data via the interconnection, called blockshipping here .

Problems and solutions

  • An interruption of the cluster interconnect must be prevented. Because from the point of view of a cluster member it cannot be differentiated whether there is an interruption or a loss on the other side. The technology seeks to fundamentally reduce the risk of an interruption through redundancy , usually through bundling (also bonding or trunking) of the network interfaces used. The split-brain problem that occurs in extreme cases can be resolved by using one or more quorums (voting disks).
  • An overload situation on the interconnect medium can suggest a failure to the cluster manager if the function of the heartbeat is impaired by excessive transfer rates over certain defined times. As discussed above, depending on the system, the cluster interconnect is also used for the transfer of application data, care must be taken to ensure that the dimensions are adequate. Bonding / trunking can also be helpful here, as the bandwidth of the interfaces used is added in the corresponding B. mode. Basically, data streams not initiated by the clusterware on the interconnect should be avoided as far as possible.

See also

Web links