Glossary

Understanding of the following terms is useful in understanding the Netra Proxy Cache Array product. Italicized terms in definitions are defined elsewhere in the glossary.

administrative host

One host in a Netra Proxy Cache Array on which you perform administrative functions and from which you propagate configuration data to all members of the array. It is recommended that the administrative host also act as DNS server for the array.

acquire message

A control message, sent by a service group leader, requesting that a service group member acquire a service address. A leader sends an acquire message when a member first joins the service group, when the new member first obtains its service address. A leader also sends an acquire message when a member host/service fails and the leader needs to reassign an orphaned service address.

acquire script

Script that is executed upon the acquisition of a service address, invoked in response to an acquire message. For successful execution, a monitor object needs to supply to an acquire script the service address being acquired and the name of the interface associated with that service.

acquiring

One of four address states of a service address. Acquiring is an intermediate state, between unserved and online. A service group leader acts on service address in only the unserved and online states. The acquiring state allows a service to perform initialization tasks; because the service is acquiring and not unserved, the leader will not reassign the initializing service's address to another host.

address states

A service address can be in one of four states: unserved, acquiring, online, and releasing. A service group leader acts on service address in only the unserved and online states. Acquiring and releasing are intermediate states between unserved and online.

cache

A store of information used for repeated fast access. With respect to web pages, a cache is a repository of URLs that lies between clients (browsers) and origin web servers. Design issues for caches involve which pages to store, for how long. Caches are useful when the ratio of hits to misses is at least 30% (approximately). This threshold varies according to the types of objects being cached and the expense and bandwidth of the links between the cache and the origin web server. For example, a hit rate of 20% might be cost effective if the hits are for very large objects. Similarly, if the links between the cache and origin web server are very expensive, even a low hit rate might prove cost effective. Another important design issue is how expensive a miss is. For example, if a cache were of such size that an exhaustive search is a resource-intensive activity, it might be more cost effective not to have the cache.

control address

IP address of a control interface. Each control interface has a control address.

control interface

The monitor object's network interface to the control network. The monitor object transmits and receive information messages over a control interface.

control message

A release message or an acquire message multicast over a service group's control network.

control network

A logical network over which service group members exchange information messages. A control network is distinguished from a service network. It might be an IP subnet distinct from or the same as the IP subnet used for the service network.

failover

The sequence of steps initiated by a service's monitor object when a host/service failure is detected. The result of these steps is that the service address associated with one host/service instantiation is moved to a different host.

hit

The finding of a requested URL in a cache, obviating the need to request the object from a parent or an origin web server. A hit is distinguished from a miss.

HTTPS

An advanced form of SSL tunneling supported by popular web browsers.

information message

A message multicast at a regular interval over a service group's control network. Each member of a service group multicasts an information message and listens for information messages sent by other members. An information message is, in itself, a heartbeat and, in addition, communicates the load and capacity of host/service instantiations and the address states of service addresses in the group.

Inter Cache Protocol (ICP)

A lightweight protocol used by a proxy cache server to inquire of its siblings and ICP-capable parents as to whether they have a web object. The ICP contains a metric (response time) that enables a requestor to choose among multiple caches.

leader

See service group leader.

load object

A script or program that measures the load and capacity of a host resource. A load object returns two integers, indicating current load and current capacity. A load object can be shared by multiple monitor objects. A given monitor object can obtain data from multiple load objects. See test object.

miss

The failure to find a requested URL in a cache, requiring further activity on the part of the cache server. This activity might include any or all of sending ICP requests to parents and siblings, sending requests to parents, and sending a request to an origin web server. A miss is distinguished from a hit. To a large degree, the success of a web cache depends on the expense associated with a miss.

monitor object

A service-specific module that is incorporated into the array daemon. A monitor object manages the service addresses assigned to a service. It performs failover, where necessary, and can address load-balancing by modifying a service group DNS zone. A monitor object is associated with one or more load objects and test objects.

neighbor

Refers to parents and siblings of a proxy cache server.

online

One of four address states of a service address. An online address is one that identifies a service that can perform work for a requesting client. In a service group where DNS is employed, only online addresses are included in the group's DNS zone. A service group leader acts on service address in only the unserved and online states.

orphaned service address

A service address not assigned to its preferred host. If the preferred host of a service address fails and releases its service address (that is, its preferred service address) that address is considered orphaned. It remains orphaned until such time as it is reassigned to its preferred host or the expiration of the service timeout, whichever occurs first.

parent

An proxy cache server that is responsible for returning a requested object if a child server cannot retrieve the object from its own cache. The parent attempts to locate the requested object in its own cache. If it cannot, it requests the object from its parents and siblings, or, in the absence of parents and siblings, from the origin web server. A parent might be ICP-capable, in which case it receives ICP queries for objects along with a proxy cache server's siblings.

persistent connections

A feature of HTTP 1.1, as implemented by popular web browsers, wherein multiple, different HTTP requests can be carried on the same TCP connection. Sometimes referred to as "HTTP keepalive".

preferred host of a service address

The host that is originally assigned a service group address. This becomes the host's preferred service address. In response to host and service failures, a service address might get moved from its preferred host. However, whenever conditions permit, array activities return a service address to its preferred host.

preferred service address

The service address associated with a service on a host when that host becomes a member of a service group. If there is a host or service failure on the host that "owns" a preferred service address, the releases that address, at which point the address becomes an orphaned service address, subject reassignment to a host other than its original owner. At such time as the original owner returns to health, it reacquires its preferred service address.

quiesced host

A host that can release, but cannot acquire service addresses; it cannot acquire its own preferred service address. A quiesced host is in a state between failed and available. As with a failed host, the service addresses of a quiesced host are not included in the service group DNS zone. Unlike a failed host, a quiesced host can continue to serve its current service addresses. A host is quiesced via the array daemon's SNMP interface or through a command-line interface. The quiesced state enables a system administrator to address resource problems on a host without removing the host from the service group.

release message

A control message, sent by a service group leader, requesting that a service group member release a service address. A leader sends a release message to a member that has been serving a previously orphaned preferred service address when the original owner of that preferred service address is ready to resume ownership. A release message invokes a release script on the host releasing the service address.

release script

A script invoked in response to a release message. A release script releases the resources associated with a service. For successful execution, a monitor object needs to supply to a release script the service address being released and the name of the interface associated with that service.

releasing

One of four address states of a service address. Releasing is an intermediate state, between unserved and online. A service group leader acts on service address in only the unserved and online states. The releasing state allows a service to perform cleanup tasks; because the service is releasing and not yet unserved, the leader will not reassign the initializing service's address to another host before cleanup is complete.

service address

The address at which a service group provides a network service to clients. A set of service addresses is the essential entity managed by the monitor objects in a service group.

service group

A set of hosts that, collectively, provide a network service to clients. Members of a service group run an array daemon that multicasts heartbeat messages. If a service or host in the group fails, the address associated with that service/host is transferred to another member of the group.

service group DNS zone

The DNS name space containing all hostname-to-IP address mappings in a service group. The monitor object can modify the contents of a DNS zone, depending on the vitality of the service group members.

service group leader

The entity within a service group that assigns preferred service addresses to member hosts and assigns orphaned service addresses to hosts capable of acquiring them. Election of the leader occurs automatically as a result of the exchange of information messages among service group members. A group always has only one leader. The role of leader moves to a different member upon failure of the leader host. The leader role might move to a different host upon addition or removal of a member host to or from the service group, depending on the result of the election algorithm.

service interface

The interface through which a host provides a service to clients. A service interface is identified by a service address.

service network

The IP subnet over which a service group provides a service to clients. See control network.

service timeout

The maximum length of time a service address can be orphaned (see orphaned service address). Beyond this period, the service group leader deinstalls the service address from the service group.

sibling

A proxy cache server that has a peer relationship with another proxy cache server. If a proxy cache server receives a request for an object that it cannot fulfill, it checks its parents and siblings to see whether they have the object. A sibling server checks only its local cache and no further. A parent checks its own cache and, if the object is not there, attempts to retrieve the object from its parent or from an origin web server.

Secure Sockets Layer (SSL) tunneling

Involves opening a direct socket between the client (browser) and a target web server for secure communication. A tunneled connection might pass through one or more proxy cache servers.

test object

A script or program that tests the functionality of a host resource. A test object might return, for a example, a boolean, indicating whether a control interface can transmit and receive. Test objects return data to monitor objects. A test object can be shared by multiple monitor objects. A given monitor object can obtain data from multiple test objects. See load object.

unserved

One of four address states of a service address. An unserved address is one that is not being served by any service group member. Under certain conditions, a service group leader attempts to assign an unserved address to a host that has the largest amount of excess capacity (determined by capacity minus the load, as returned by a host's load object). A service group leader acts on service addresses in only the unserved and online states.

web object

A web page, audio or video clip, graphic file, or other object that can be provided by a web server to a client (most often a browser) using the HyperText Transfer Protocol. A proxy cache server caches web objects, although not all proxy cache servers can cache all types of web objects.