|Oracle® Coherence Developer's Guide
Part Number E18677-01
The following sections are included in this chapter:
A Coherence cluster is a collection of JVM processes. At run time, JVM processes that run Coherence automatically join and cluster. JVMs that join a cluster are called cluster members or cluster nodes. Cluster members communicate using Tangosol Cluster Management Protocol (TCMP). Cluster members use TCMP for both multicast communication (broadcast) and unicast communication (point-to-point communication).
A cluster contains services that are shared by all cluster members. The services include connectivity services (such as the Cluster service), cache services (such as the Distributed Cache service), and processing services (such as the invocation service). Each cluster member can provide and consume such services. The first cluster member is referred to as the senior member and typically starts the core services that are required to create the cluster. If the senior member of the cluster is shutdown, another cluster member assumes the senior member role.
TCMP is an IP-based protocol that is used to discover cluster members, manage the cluster, provision services, and transmit data. TCMP uses a combination of UDP/IP multicast, UDP/IP unicast and TCP/IP as follows:
Cluster discovery: Multicast is used to discover if there is a cluster running that a new member can join.
Cluster heartbeat: The most senior member in the cluster issues a periodic heartbeat through multicast; the rate can be configured and defaults to one per second.
Message delivery: Messages that must be delivered to multiple cluster members are often sent through multicast, instead of unicasting the message one time to each member.
Direct member-to-member ("point-to-point") communication, including messages, asynchronous acknowledgments (ACKs), asynchronous negative acknowledgments (NACKs) and peer-to-peer heartbeats.
Under some circumstances, a message may be sent through unicast even if the message is directed to multiple members. This is done to shape traffic flow and to reduce CPU load in very large clusters.
An optional TCP/IP ring is used as an additional "death detection" mechanism, to differentiate between actual node failure and an unresponsive node, such as when a JVM conducts a full GC.
TCP/IP is not used as a data transfer mechanism due to the intrinsic overhead of the protocol and its synchronous nature.
The TCMP protocol provides fully reliable, in-order delivery of all messages. Since the underlying UDP/IP protocol does not provide for either reliable or in-order delivery, TCMP uses a queued, fully asynchronous ACK- and NACK-based mechanism for reliable delivery of messages, with unique integral identity for guaranteed ordering of messages.
The TCMP protocol requires only three UDP/IP sockets (one multicast, two unicast) and six threads per JVM, regardless of the cluster size. This is a key element in the scalability of Coherence, in that regardless of the number of servers, each node in the cluster can still communicate either point-to-point or with collections of cluster members without requiring additional network connections.
The optional TCP/IP ring uses a few additional TCP/IP sockets, and a total of one additional thread.
The TCMP protocol is very tunable to take advantage of specific network topologies, or to add tolerance for low-bandwidth and high-latency segments in a geographically distributed cluster. Coherence comes with a pre-set configuration. Some TCMP attributes are dynamically self-configuring at run time, but can also be overridden and locked down for deployment purposes.
Coherence functionality is based on the concept of cluster services. Each cluster node can participate in (which implies both the ability to provide and to consume) any number of named services. These named services may exist, which is to say that they may be running on one or more other cluster nodes, or a cluster node can register new named services. Each named service has a service name that uniquely identifies the service within the cluster, and a service type, which defines what the service can do. There may be multiple named instances of each service type (other than the root Cluster service). By way of analogy, a service instance corresponds roughly to a database schema, and for data services, a hosted named cache corresponds roughly to a database table. While services can be added, many applications only require the default set of services shipped with Coherence. There are several service types that are supported by Coherence.
Cluster Service: This service is automatically started when a cluster node must join the cluster; each cluster node always has exactly one service of this type running. This service is responsible for the detection of other cluster nodes, for detecting the failure (death) of a cluster node, and for registering the availability of other services in the cluster. In other words, the Cluster Service keeps track of the membership and services in the cluster.
Proxy Service: While many applications are configured so that all clients are also cluster members, there are use cases where it is desirable to have clients running outside the cluster, especially in cases where there are hundreds or thousands of client processes, where the clients are not running on the Java platform, or where a greater degree of coupling is desired. This service allows connections (using TCP) from clients that run outside the cluster.
Invocation Service: This service provides clustered invocation and supports grid computing architectures. Using the Invocation Service, application code can invoke agents on any node in the cluster, or any group of nodes, or across the entire cluster. The agent invocations can be request/response, fire and forget, or an asynchronous user-definable model.
Distributed Cache Service: This is the distributed cache service, which allows cluster nodes to distribute (partition) data across the cluster so that each piece of data in the cache is managed (held) by only one cluster node. The Distributed Cache Service supports pessimistic locking. Additionally, to support failover without any data loss, the service can be configured so that each piece of data is backed up by one or more other cluster nodes. Lastly, some cluster nodes can be configured to hold no data at all; this is useful, for example, to limit the Java heap size of an application server process, by setting the application server processes to not hold any distributed data, and by running additional cache server JVMs to provide the distributed cache storage. For more information on distributed caches, see "Distributed Cache".
Replicated Cache Service: This is the synchronized replicated cache service, which fully replicates all of its data to all cluster nodes that run the service. Furthermore, it supports pessimistic locking so that data can be modified in a cluster without encountering the classic missing update problem. With the introduction of near caching and continuous query caching, most of the functionality of replicated caches is available on top of the Distributed cache service (and with better robustness). But replicated caches are often used to manage internal application metadata. For more information on replicated caches, see "Replicated Cache".
Optimistic Cache Service: This is the optimistic-concurrency version of the Replicated Cache Service, which fully replicates all of its data to all cluster nodes, and employs an optimization similar to optimistic database locking to maintain coherency. Coherency refers to the fact that all servers end up with the same "current" value, even if multiple updates occur at the same exact time from different servers. The Optimistic Cache Service does not support pessimistic locking, so in general it should only be used for caching "most recently known" values for read-only uses. This service is rarely used. For more information on optimistic caches, see "Optimistic Cache".
Regarding resources, a clustered service typically uses one daemon thread, and optionally has a thread pool that can be configured to provide the service with additional processing bandwidth. For example, the invocation service and the distributed cache service both fully support thread pooling to accelerate database load operations, parallel distributed queries, and agent invocations.
It is important to note that these are only the basic clustered services, and not the full set of types of caches provided by Coherence. By combining clustered services with cache features such as backing maps and overflow maps, Coherence can provide an extremely flexible, configurable, and powerful set of options for clustered applications. For example, the Near Cache functionality uses a Distributed Cache as one of its components.
Within a cache service, there exists any number of named caches. A named cache provides the standard JCache API, which is based on the Java collections API for key-value pairs, known as
Map interface is the same API that is implemented by the Java
Hashtable class, for example.