Method and apparatus for administering a server having a subsystem in communication with an event channel6789112Abstract Methods and apparatus for administering a remote server having a subsystem in communication with an event bus. In one aspect of the present invention, an administration tool for administering a server has a subsystem in communication with an event bus. The administration tool includes a graphical user interface communications channel and a graphical user interface module corresponding to the server subsystem, wherein the graphical user interface module is in communication with the channel. The administration tool also includes a transport module in communication with the channel and the graphical user interface module. The graphical user interface module transmits an administration command to the corresponding server subsystem by sending the command to the transport module via the communications channel. Claims What is claimed is: Description FIELD OF THE INVENTION
TABLE 1
Parent Node ID Node ID Node Name
none 0 Root (implied)
0 1 farm_1
0 2 farm_2
1 3 Authorized_users
2 4 Authorized_users
3 5 user_1
4 6 user_1
To avoid having to traverse the entire tree upon each access to an object in the persistent store 230, a requesting subsystem 300 can dynamically bind to a particular node in the tree to serve as a starting point for traversing the tree. The particular node in the tree depends upon the type of subsystem. Generally, each subsystem 300 owns part of the tree, that is, the subsystem owns those objects that it stored in the tree. Thus, the particular node can operate as a root node for objects that the subsystem owns and a starting point from which to traverse those objects. For example using Table 1, a subsystem 300 can bind to the authorized_users node to serve as a starting point for searching for a particular user. As an illustrative example, consider that the administration tool 140 wants to authenticate whether a remote user of the server farm 110 is authorized to access an application program on a particular server 180 that is part of that server farm 110. The administration tool 140 directs an administration subsystem (not shown) to send an event message to the persistent store system service module 352 via the service locator 354 and the event delivery object 312 to obtain the desired information. The persistent store system service module 352 receives and parses the event message to obtain the distinguished name of the entry (described below) and attributes that are being requested. The format of the distinguished name corresponds to the external model used by the administration subsystem when forming the event message. An example of such a distinguished name is "root/farm_name/authorized_users/user_1." Assuming that the contents of the persistent store 230 are organized into a single tree, the persistent store system service module 352 traverses the tree to obtain information about the authorized users of that particular application. The persistent store system service module 352 traverses "down" the tree to determine whether the last node traversed matches the distinguished name (in this case, whether the user_1 is included as an authorized user). In this manner, as long as the distinguished name in the external model maintains a hierarchical order that corresponds to a tree structure (internal model) in the persistent store 230, the individual/arbitrary formats of each element of the distinguished name do not need to be analyzed. Data ownership and security issues are also important considerations when sharing a common persistent storage environment across multiple subsystems (requesting entities). The subsystem 300, which is the source of the data, sets the access restrictions via the SAL API of the persistent store system service module 352 that limit the exposure of the data to an authorized subset of requesting entities via the SAL API. 3.5 The Dynamic Store System Service Module The dynamic store 240 operates as a global database that stores records accessible by each server 180 in a zone 260, 270. In one embodiment, each stored record is an attribute-value pair. An example of an attribute is subsystem identifier; an example of a value is the actual subsystem ID number. Each subsystem 300 that uses the dynamic store 240 defines the schema of the records that are created and stored for that subsystem type. Different subsystems generally have different schemas. The first call that a subsystem 300 makes to the dynamic store 240 registers the schema that subsystem will use. Subsequently, all subsystems 300 of the same subsystem type that register with the dynamic store 240 can access the records created according to that registered schema. As part of registering the schema, a subsystem 300 can specify which attributes may be used for searching. In one embodiment, a subsystem 300 identifies one or more attributes that will be frequently used to search the record table. In one embodiment, each record is stored by both the server 180 creating the record as well as the server 180 responsible for storing records of that type. For embodiments in which more than one zone 260, 270 exists in a farm 110, a record is stored on a server 180 in each zone 260, 270 identified by the zone master of each zone as the server 180 that stores records of that type. The server 180 creating the record essentially acts as redundant storage for the table record. In some embodiments, the table owner updates the server 180 creating the record with subsequent changes to the record. Within a zone 260, 270 the definitive authority as to the correct value of a table record is the table owner, i.e., the server 180 chosen by the zone master to store data records of that type. Between zones, the definitive authority as to the correct value of a table record is the table owner in the zone from which the record originated. Although there are definitive authorities as to the correct value for a table record, no definitive authority exists as to the contents of a table--a table's contents are the union of all table records stored throughout the farm 110. Each server 180 in the server farm 110 has a dynamic store system service module 356 that handles all calls from subsystems 300 to the dynamic store 240. The dynamic store system service module 356 permits each subsystem to perform database operations on the dynamic store 240. The operations are: (1) to insert a record, (2) to delete a record, (3) to search the dynamic store 240 to retrieve all records satisfying certain specified criteria, and (4) to update one or more values for attributes in an existing record. When a record is inserted into a table or when a record is updated, the server 180 requesting the change locally stores the record and forwards it to the owner of the table. The name of the server changing or creating the record can be added as an attribute to the record to facilitate informing that server of subsequent changes to the record that may be effected by other servers 180 in the farm 110. The requesting server 180 uses its local copy of the record if the table owner changes unexpectedly, for example, if the table owner crashes. When the zone manager detects this problem and designates a new table owner, the servers 180 in the server farm 110 upload locally-stored table records to the new owner. Records can be queried based on attribute, and any number of records may be returned from a query. When a server 180 receives a query request, it forwards the request to the table owner, which performs the search and returns the results. The server that originated the query may cache the search results depending on various criteria such as configuration or record consistency parameters. The delete operation is similar to a query, in that any valid search parameters can be used to specify which records to delete. This allows for operations such as "delete all records from host ABC." Just as with a query request, the delete request is forwarded to the appropriate table owner. Since some of the records being deleted may have been created on the requesting server, the table owner returns a list of the records that were actually deleted. This allows the local server 180 to delete locally-stored records. In one embodiment, when a subsystem 300 registers its schema (i.e., defines the data structure) with the dynamic store 240, that subsystem 300 also supplies one or more parameters that specifies usage information about records. One such parameter controls "update latency," that is, the frequency at which the records are updated. Every subsystem 300 on every server 180 can independently determine this frequency and therefore every server 180 in the server farm 110 can see the same information in the records associated with that subsystem 300. Another parameter is the "time to live after originating host is no longer present." This parameter is useful for maintaining the record although the originator of the record is no longer active. When the time to live is set to zero, the record is deleted immediately after the absence of the originating host is detected by the record owner, i.e., the collector point responsible for collecting records of that type. The record owner is the only subsystem entitled to delete this record. Yet another parameter is a "time to live" parameter that results in automatic deletion of a record by the dynamic store system service module 356 when the "time to live" is exceeded. Time starts from the insertion of that record into the dynamic store 240. Through communication among the servers in the server farm 110, there is a dynamic election of a master server in every zone defined in the server farm 110. After the master server is elected, all other servers in the zone know the identity of the master server, as described in more detail below. At least one copy of every record in the dynamic store 240 exists in each zone. In one embodiment, the master server of the zone stores every record in memory local to that master server. In another embodiment, the master server distributes the dynamic store 240 in the local memory 325 of some or all of the servers 180 in the zone based on record type. The determined server is thus designated as the collector point for that record type. Should one of the servers in the server farm fail, the master server chooses a new server in the zone to hold the type of records that the failed server previously held. This new server requests an update of those records from every other server in that zone to replace the records that became inaccessible when the server failed. Because every server keeps a copy of the records that pertain to that server, the update restores the content of the dynamic store 240. If the master server fails, any server in the zone that detects the absence of the master server initiates an election for a new master server. In one embodiment, master servers are the only servers that know the master servers of the other zones 260, 270. To obtain this information, each master server queries every server in each other zone 260, 270, seeking a response that identifies the master server of that zone 260, 270. Zones are preconfigured, and the identity of servers associated with zones 260, 270 is stored in the persistent store 230. Periodically, each master server of a zone 260, 270 sends the records in the dynamic store 240 for that zone 260, 270 to the master servers in the other zones 260, 270. In another embodiment, each server that holds the records sends a copy of those records to corresponding servers in the other zones 260, 270. Such servers determine who are the corresponding servers in the other zones 260, 270 from information collected by the master server of its own zone 260, 270. 3.6 Service Locator System Service Module Referring again to FIG. 3, the service locator 354 is in communication with each subsystem 300 over the event bus 310 (or via its internal API). The service locator 354 identifies a server 180 for servicing events issued to other subsystems 300. The identified server 180 can be local or remote. In brief overview, a source subsystem 300 may create or issue an event for which the host of the destination subsystem is not determined before the source subsystem 300 issues the event. In these cases, the source subsystem 300 uses either a SAL API call or an internal API call provided by the service locator 354 to either (1) obtain the address of the server 180 hosting the destination subsystem 300 or (2) request that the service locator 354 deliver an event to the destination subsystem 300 on behalf of the source subsystem 300. The service locator 354 identifies a destination host by accessing information maintained in the dynamic store 240 through the dynamic store system service module 356 (see section 3.5). This information provides a zone-wide inventory of the server components in the server farm 110; that is, the information indicates which subsystems (and the versions of those subsystems) are installed on every server 180 in the server zone 260, 270. This information also indicates which of such servers 180 in the zone 260, 270 are currently operating. Thus, through this information, the service locator 354 has knowledge of all available subsystems 300 in the zone 260, 270. Every server 180 in the server farm 110 has a service locator 354 that contributes to the zone-wide information in the dynamic store 240. For example, when a server 180 becomes operational, each subsystem 300 installed on the server 180 registers with the service locator 354. In one embodiment, the service locator 354 provides a "RegisterService" function that may be called by a subsystem (either through the SAL API or the internal API of the service locator 354) in order to register services that it can provide to other subsystems. In one embodiment, subsystems 300 register with the service locator 354 each version of each event that the subsystem 300 will process. In another embodiment, the RegisterService function also accepts as a parameter a rank value, which indicates the relative importance of the subsystem 300. Upon receipt of the registration message, the service locator 354 makes an entry into the dynamic store 240 for that subsystem 300. The entry includes the information provided by the subsystem, such as its identifier and its rank, when provided. Table 2 below depicts one embodiment of a table stored in the dynamic store 240.
TABLE 2
Subsystem
ID Rank Zone Host ID
FFFF 1 A 0015
AAAA 0 A 0012
FFFF 1 A 0009
AAAA 0 A 0006
When a server 180 shuts down in a controlled fashion, it is removed from the zone 260, 270, and an "UnregisterService" call is made to the service locator 354 by each subsystem 300 resident on that server 180. This call informs the service locator 354 that those subsystems are no longer present in the zone 260, 270. In some embodiments, the service locator 354 instructs the dynamic store 240 to discard records associated with a server 180 that terminates execution unnaturally, e.g., crashes. To determine the target host for servicing an event, the service locator 354 determines certain information: (1) which servers 180 host the type of subsystem 300 identified in the event as the destination subsystem, and (2) which of such servers 180 is the target host for handling the event. Upon determining the target host, the service locator 354 either returns the determined address to the requesting subsystem 300 or it modifies a received event to include the determined address as the addressing information for the event and it delivers the modified event to the event bus 310 for delivery to that host. Referring back to Table 2, an embodiment of a table stored in the dynamic store 240 by service locators 354 is shown that includes entries for two subsystems (having identifiers FFFF and AAAA). Each entry includes a subsystem identifier, a subsystem rank, a zone identifier, and a host identifier. The service locator 354 receives a request for an address (or a request to deliver an event to a host) and accesses the table stored in the dynamic store 240. In some embodiments, the service locator 354 provides two function calls that return a target host identifier to the requesting subsystem 300: "GetBestHost," which returns the host identifier associated with a host that can handle a particular type of event; and "GetBestHostFromList," which returns a target host identifier selected from an input list of hosts. If the table has only one entry for which the subsystem identifier matches the subsystem identifier provided in the API call, the host identifier from that table entry is returned to the requesting subsystem 300. If more than one table entry has a matching subsystem identifier, i.e., there is more than one host in the zone that can process the subject event, a host identifier is selected based using a predetermined rule or set of rules. For example, a host identifier may be selected at random, in round-robin order, based on the rank associated with the table entry, or based on other information that may be stored in table such as network latency to host, available bandwidth of channel between requesting subsystem 300 and target host, or geographic proximity to the requesting subsystem 300. The service locator 354 may also provide API calls for sending an event to the target host on behalf of the requesting subsystem 300. In these embodiments, if only one of the other servers in the zone can process the identified message, i.e., there is only one entry in the table, then the service locator 354 inserts the host identification of that server into the event and sends the modified event to the event bus 310 for delivery to the target host. If more than one other server in the zone has the destination subsystem, then the service locator 354 chooses one of the servers using any of a variety of criteria as described above, modifies the event as described above, and transmits the modified event to the target host. Using Table 2 as a specific example, a subsystem 300 may issue a GetBestHost call for a subsystem having an identifier of "FFFF." Two servers host that subsystem, identified by an identifier of 9 and 15. The identifier corresponding to either of these hosts may be returned to the requesting subsystem. In one embodiment, the system administrator can force one of the two subsystems to be elected by changing the "rank" values in the table. For example, if the entry associated with host "15" has a higher rank than the entry associated with host "9," host "15" may always be selected as the target host. 3.7 Subscription Manager System Service Module The subscription manager 362 manages subscriptions for a server 180. A subscription is a standing request by which a subscribing subsystem 300 publicizes to the subscription manager 362 of the local server and/or to the subscription managers of remote servers that the subscribing subsystem wants to be notified upon the occurrence of an event. The registered subscription identifies the event and the subscribed-to subsystem that produces the event. Upon the occurrence of that event, the subscription manager 362 sends the event to any subsystem that has registered a subscription to that event by way of the event delivery object 312. The subscription manager 362 uses two tables for managing subscriptions: (1) a local subscription table 450, and (2) a remote subscription table 418. 3.7.1 Local Subscription Table The local subscription table 450 resides in local server memory 325 and stores subscriptions for which the specified scope is local. Using the local subscription table 450, the subscription manager 362 can alert local subsystems 300 of the occurrence of particular events on the server 180. Any local subsystem 300 on any server 180 can request to be notified when a particular subsystem issues a particular event by posting a subscription for that occurrence in the local subscription table 450. Referring to FIG. 4B, and in more detail, the local subscription table 450 includes an entry 460 for each posted subscription. In one embodiment, each entry 460 of the local subscription table 450 includes event field 462 identifying a unique event, a subsystem field 464 identifying the subsystem that owns (i.e., generates) the unique event, and a destination subsystem field 468 identifying the subsystem 300 subscribing to the unique event. An exemplary local subscription is illustrated in FIG. 4B in which subsystem 300 seeks to be notified when subsystem 300' posts an "I'm Up" event to the event delivery object 312. For purposes of illustrating this subscription, names corresponding to the subsystem 300' and the service locator 354 appear in the fields 464 and 468, respectively, but the actual implementation of this subscription can use pointers to such subsystem 300' and service locator 354. 3.7.2 Remote Subscription Table A remote subscription table 480 is stored in the dynamic store 240 and stores subscriptions registered by specific remote servers or having a scope specified as zone or farm-wide. Placing such subscriptions in the dynamic store 240 makes the subscriptions accessible farm-wide by subscription managers 362 of every other server 180 in the server farm 110. In one embodiment, shown in FIG. 4C, the remote subscription table 480 is implemented as three separate tables: a first table 480' stores subscriptions to events that may occur in the same "zone," a second table 480" stores subscriptions to events that may occur anywhere in the server farm 110, and a third table 480" stores subscriptions to events that may occur on a specifically identified remote host. In more detail, each table 480', 480", and 480'" (generally 480) includes an entry 484 for each posted subscription. In one embodiment, each entry 484 includes an event field 492 identifying a unique event, a subsystem field 494 identifying the subsystem that owns (i.e., generates) the unique event, a destination subsystem field 496 identifying the subsystem 300 subscribing to the unique event, and a subscribing host field 498 identifying the host of the subscribing subsystem. The table 480'" further includes a source host identifier 488 for identifying the specific remote host upon which the subscribed-to subsystem resides. An exemplary subscription is illustrated in FIG. 4C in which subsystem 300 seeks to be notified when subsystem 300' of a particular remote host server 180' posts an "I'm Up" event. For purposes of illustrating this subscription, which is placed in the specific remote table 480'" of the remote subscription table 480, names corresponding to the servers 180, 180' and subsystems 300', 300 appear in the entry 484, but the actual implementation of this subscription can use pointers to such servers 180, 180' and subsystems 300', 300. The subscription manager 362 provides three functions that can be called by other subsystems 300: (1) Subscribe, (2) Unsubscribe, and (3) PostNotificationEvent. In one embodiment, these functions are called through the SAL 304 associated with the subscription manager 362. In another embodiment, the functions are called through the internal API provided by each subscription manager 362. 3.7.3 Subscribe Function When a subsystem 300 wants to subscribe to an event of another subsystem 300, the subscribing subsystem 300 calls the Subscribe function (either via a SAL API call or an internal API call) provided by the subscription manager 362. The Subscribe function instructs the subscription manager 362 to register a subscription in either the local subscription table 450 or in the remote subscription table 480 held in the dynamic store 240. The subscribing subsystem 300 specifies the scope of the subscription: local, zone, or farm-wide. In one embodiment, the specific SAL call used by the subscribing subsystem 300 determines the scope of the subscription. In another embodiment, the scope is an input parameter of the SAL call. The event delivery object 312 of the event bus 310 dispatches the Subscribe event to the subscription manager 362. Typically, those subsystems 300 that are initialized after the subscription manager 362 is initialized call the Subscribe function during the initialization of such subsystems 300. The Subscribe function can also be called anytime during server operation by any subsystem. Input parameters to the Subscribe function uniquely identify the subscribing subsystem, the event for which the subscribing subsystem requests notification, the subscribed subsystem to be monitored, and, optionally, the scope of the subscription. In one embodiment, the parameters uniquely identifying the subscribing and subscribed subsystems 300 may each be implemented as two separate entities: a value identifying the subsystem 300 and a value identifying the host on which the subsystem 300 resides. In other embodiments, the Subscribe function returns an output value representing the status of the subscription request, such as successfully registered. Upon receiving the Subscribe function call, the subscription manager 362 determines the scope of the subscription from the type of SAL call 304 used to deliver the Subscribe event. If the scope of the subscription is for a local subsystem, then the subscription manager 362 stores a corresponding subscription entry in the local subscription table 450. If the scope of the subscription is remote, the subscription manager 362 communicates with the dynamic store subsystem 370 over the event bus 310 to register the subscription in the appropriate section of the remote subscription table 480 in dynamic store 240. 3.7.4 Unsubscribe Function A subscribing system 300 can remove a previously registered subscription from the local and remote subscription tables 450, 480 by issuing an Unsubscribe function to the subscription manager 362. Such subscribing subsystem 300 can unsubscribe to only those subscriptions that the subsystem 300 has previously registered. Input parameters to the Unsubscribe function uniquely identify the subsystem requesting removal of the subscription, the event for which the subscribing subsystem no longer requests notification, and the subsystem having the subscription to be removed. The input parameters that uniquely identify the subscribing and subscribed-to subsystems are implemented in one embodiment as two separate entities: a value identifying the subsystem and a value identifying the host on which that subsystem resides. In response to an Unsubscribe function call, the subscription manager 362 searches the local subscription table 450 and remote subscription tables 480 and removes every entry corresponding to the subscription to be removed. To remove the subscription from the remote subscription tables 480, the subscription manager 362 sends a delete request to the dynamic store system service module 356 to remove the entries from the dynamic store 240. The Unsubscribe function returns an output value representing the status of the removal of the subscription, such as successfully completed. 3.7.5 PostNotificationEvent Some subsystems 300 produce events that may be subscribed to by other subsystems that are local and/or remote to these subsystems. Upon issuing such an event, such subsystems 300 also call a PostNotficationEvent function to send a copy of this event to the subscription manager 362. The subscription manager 362 issues a copy of that event to local or remote subscribing subsystems 300. The subsystems 300 call the PostNotificationEvent function regardless of whether any subsystem has actually registered a subscription to that event, because only the subscription manager knows if an event has been subscribed to by another subsystem. FIG. 5A shows an embodiment of a process used by the subscription manager 362 upon receiving (step 510) a Subscribe function command. From the event type, the subscription manager 362 determines (step 514) whether the scope of the subscription event is remote. If the subscription is not remote in scope, the subscription manager 362 stores (step 518) the subscription in the local subscription table 450. When the scope of the subscription is remote, the subscription manager 362 determines (step 522) whether the subscribed-to event is in the zone, farm-wide, or for a specific remote host. Then the subscription manager 362 inserts (step 526) the subscription into the appropriate table 480', 480", 480'". The inserted subscription (hereafter, a subscription record) follows the particular schema defined by the subscription manager 362. A similar process is used to remove subscriptions from the subscription tables 450 and 480 upon receiving an Unsubscribe call. FIG. 5B shows an embodiment of a process used by the subscription manager 362 for each PostNotificationEvent received (step 550) by the subscription manager 362. The subscription manager 362 determines (step 554) if the event exists in the local subscription table 450. If the event is subscribed to by one or more local subsystems, then the subscription manager 362 generates (step 558) a copy of the event to be delivered to each subscribing local subsystem. Each copy of the event is placed in its own event buffer 380. Then the subscription manager 362 checks (step 562) the zone table 480' for any subscribing servers in the same zone. Similarly, the subscription manager 362 requests searches (steps 566 and 570) for subscriptions in the farm-wide section 480" and specific remote host section 480'", respectively, of the remote subscription table 480. In one embodiment, for each access to the remote subscription tables 480, the subscription manager 362 issues an event to the dynamic store system service module 356 that causes the desired search. Then, in one embodiment, rather than search the local dynamic store 240 directly, the subscription manager 362 sends a copy of the event to a subscription dispatcher. The subscription dispatcher is one of the servers 180 in the server farm 110 that is dedicated for dispatching events to remote subscribers (i.e., another server in the same or different zone). The subscription dispatcher is identified as the target host in the zone for handling subscribed-to events. For each received event, the subscription dispatcher performs a search operation on the remote subscription tables 480 in the dynamic store 240 and retrieves all subscription records corresponding to subscribers of that event. Each retrieved subscription record corresponds to one subscription. The subscription manager 362 then produces an event for each retrieved record, inserting the identification of the subscribing subsystem into the appropriate field in that event. 3.8 Host Resolution System Service Module A subsystem 300 may target events to another subsystem residing on a remote server. Parameters associated with issuing such events include a unique host identifier corresponding to the remote server. The host resolver 360 receives such events from these source subsystems 300 (and in other embodiments from other system service modules 350) requesting that a distinguished name be obtained for the remote server. To obtain the distinguished name, the host resolver 360 sends an event that includes the unique host identifier to the persistent store system service module 352. The persistent store system service module 352 uses the unique host identifier to search the persistent store 230 for a corresponding distinguished name, and returns the distinguished name and the port address to the host resolver 360. The host resolver 360 can return the distinguished name and port address to the source subsystem 300 or it may forward the event received from the source subsystem 300 to the host identified by the distinguished name on behalf the source subsystem 300. 3.9 Zone Manager System Service Module Each server 180 in the server farm 110 includes a zone manager 358 that directs accesses to the dynamic store 240 made by the dynamic store system service module 356 to the server 180 responsible for collecting data of the type identified in the access. One of the zone managers 358 in a server farm 110 is elected by its peers to be the master of the server farm 180. When acting as a master, a zone manager 358 (1) determines which server 180 collects each type of data, (2) designates which servers 180 in the farm 110 are responsible for providing various network services, and (3) identifies the zone master of other zones 260, 270 in the farm 110. As described above, the dynamic store 240 may be distributed among more than one server 180 in a server farm 110. 3.9.1 Assigning Ownership of Distributed Resources The dynamic store 240, in one embodiment, comprises one or more record tables managed by the dynamic store system service module 356. Record tables store information relating to server farm run-time data, such as dynamic subscription tables and disconnected sessions. The dynamic store system service module 356 queries the zone master to determine which server 180 in the zone 260, 270 stores the various record tables. The dynamic store system service module 356 can use the services of the zone master through a zone master interface, which in one embodiment provides a service called GetZoneResourceOwner. This service accepts as input a unique string identifier of an arbitrary resource, and returns the identity of the server 180 that should own a given resource. The dynamic store 230 is thus able to call GetZoneResourceOwner, passing the name of the dynamic store record table whose owner is desired, and the zone master will return the identity of the server 180 that owns that resource, i.e., that stores the dynamic store 230 records for that resource. In further embodiments, the zone master chooses which server 180 in a server farm 110 stores dynamic store record tables. In these embodiments, the zone manager may choose a server 180 based on physical characteristics, such as available memory, or other criteria, such as proximity to (either logically or physically) those entities requesting the dynamic store records. In other of these embodiments, the zone master may change which server 180 stores the record table during server farm operation. 3.9.2 Assignment Ownership of Network Services In some embodiments, certain services provided by service modules may be centralized, to allow all of the servers 180 in a zone 260, 270 make service request directly to the same zone server. An example of this might be a licensing server. In this example, all requests for a license would be directed to a single server 180 in the zone 180. The service locator system service module 354 tracks which services are available on which servers 180 in the zone 260, 270. Although in one embodiment the main purpose of the service locator system service module 354 is to find the `best` host for a given service that may be available on many servers 180 in the zone 260, 270, it is also responsible for sending messages to centralized service modules. The determination as to which of the zone's member servers should be responsible for handling a given centralized service is made by the zone master in a similar way to how it assigns ownership of zone resources. Thus, the service locator system service module 354 the zone master to determine where requests for such services should be directed. A master election can occur when a new server is added to a zone 260, 270. Alternatively, any zone manager 358 can initiate an election if the master fails to respond to a query, i.e., the master has failed. In one embodiment, any zone manager 358 may force an election at any time by broadcasting a request election event. The election results are determined by a comparison of the set of election criteria which is transmitted within the request election event transmitted by the requesting zone manager 358 with the set of election criteria maintained on each receiving zone manager 358. That is, the first election criterion from the event of the requesting zone manager 358 is compared by the receiving zone manager 358 to the first criterion of the receiving zone manager 358. The highest ranking of the two criteria being compared wins the comparison and the zone manager 358 with that criterion wins the election. If the two criteria tie, then the next criteria are sequentially compared until the tie is broken. Election criteria may be whether or not the zone manager 358 is statically configured as a master; whether the zone manager 358 is resident on the longest running server; and whether the server on which the zone manager 358 is resident has a lexically lower network name. The interaction of zone manager system service and the dynamic store system service modules 358, 356 to manage and access the dynamic store 240 is discussed in greater detail below (see section 6.3). 3.10 System Module The system module 320 is an executable program (.exe) that manages the boot-up of the server 180. Like each subsystem 300, the system module 320 is addressable (i.e., can be the target of an event) and includes an event queue 324 to receive events, such as "SetListeningPort," which sets the transport protocol port address on which the transport layer 260 "listens" for communication events. Another example of an event that can be directed to the system module 320 is "LoadSubsystem," which instructs the system module 320 to load a subsystem. Upon execution, the system module 320 initializes the event delivery object 312, the transport layer 318, and the loader module 330. The system module 320 also binds the transport layer 318 to the event delivery object 312. In one embodiment, the system module is provided as a WINDOWS NT service. In another embodiment, system module 320 is provided as a Unix daemon. 3.11 Loader The loader module 330 allows for customization of the event bus 310 for different platforms and applications. The loader 330 can be implemented as a C++ class, implemented as static code or as a dynamically linked library. In brief overview, the loader module 330 uses several functions to manage the subsystems 300. In general, the functions performed by the loader module 330 create and destroy subsystems 300. Operation of the loader module 330 is described in more detail in connection with FIG. 4. The loader module 330 uses a create function, having as input a subsystem identifier, to generate an instance of each subsystem 300. For embodiments in which an event queue is associated with the subsystem 300, the create function invokes an instantiation of an event queue in the event delivery object 312 and the loader 330 binds the event queue to the discovered subsystem 300. In other embodiments, the subsystem 300 is identified by a pointer that is entered in the dispatch table 316 to identify the subsystem 300. The event delivery object 312 uses the pointer stored in the event delivery object 312 (in some embodiments the pointer identifies an event queue) to send events to the subsystem. The subsystem 300 uses a pointer to the event delivery object 312 to deliver event to the event bus 310. Thus, for example, in embodiments in which the interfaces are provided as C++ classes, the pointers identify the desired classes. In some embodiments, this function can return a status value. The loader module 330 uses a destroy function to delete an instance of the subsystem 300 (together with an event queue, if provided, associated with that deleted subsystem) and the corresponding entry in the dispatch table 316. 4.0 Server and Subsystem Initialization FIG. 6 illustrates an embodiment of a process used to initialize a server 180, including system service modules 350 and personality subsystems 300. A server 180 executes boot service code (i.e., the system module 320) that creates the event bus 310. In the embodiment shown in FIG. 6, creation of the event bus 310 includes the steps of creating an event delivery object 312 (step 604), creating a transport mechanism 318 (step 608), and binding the event delivery object 312 to the transport layer 318 (step 612). The system module 320 instantiates a loader module 330 (step 616) and starts (step 620) execution of the loader module 330. The loader module 330 creates and loads (step 624) a specialized subsystem identified by an initialization resource. In some embodiments, the specialized subsystem is identified by an entry in a registry file. For embodiments in which system service modules 350 are provided as subsystems, the specialized subsystem instructs the loader module 330 to create and load all required system service modules 350 (step 628). The specialized subsystem also determines which personality subsystems 300 should be loaded for the server 180 (step 632). In one embodiment, the specialized subsystem accesses a registry file to determine which personality subsystems 300 should be loaded and the registry file specifies an order in which the personality subsystems are loaded. For embodiments in which the system service modules 350 are provided as subsystems, the registry file also specifies the order in which they are initialized. In one particular embodiment, the registry file specifies the following order: the persistent storage system service module 352, the dynamic store system service module 356, the zone manager 358, the host resolver 360, the service locator 354, the subscription manager 362. In another embodiment, the specialized subsystem accesses an initialization file to determine which subsystems should be loaded. In still other embodiments, the specialized subsystem accesses the persistent store 230 to determine which subsystems should be loaded. As part of loading the subsystems 300, the loader module 330 populates (step 636) the dispatch table 316 with entries 420 that map subsystem entry points to subsystem identifiers associated with the loaded subsystems 300, as shown above in FIG. 4A. Each subsystem 300 can be represented by an entry in the initialization resource, i.e. installed on the server 180, because (1) the subsystem is necessary to the operation of the server 180, or (2) the subsystem is anticipated to be useful. In one embodiment, another reason for installing a subsystem 300 is that the subsystem 300 is requested by the arrival of an event directed to that subsystem (i.e., on-demand). For such embodiments that implement on-demand loading, the loader module 330 waits until an event is received directed to that subsystem before creating that subsystem. In these embodiments, the loader module 330 provides an API that allows the loader module 330 to be invoked during run-time to create and initialize a personality subsystem 300. 5.0 Events FIG. 7A depicts an embodiment of an event 700 that includes an event header 710 and event data 730. The event header 710 is sometimes referred to as "control data" and event data 730 may be referred to as "payload data." Referring now to FIG. 7B, the event header 710 includes one or more data fields that indicate various attributes associated with the event 700. For example, the event header 710 may include: a unique event identifier (event UID) 712; an event header version identifier 714; an event data version identifier 716; an event data size indicator 718; an event data offset identifier 720; a unique identifier (UID) 722 identifying a source subsystem; a destination subsystem UID 724 identifying a destination subsystem; and a channel identifier 726, described in detail below. In more detail, the event UID 712 uniquely identifies each event produced by the subsystems 300 and system service modules 350. Every subsystem and system service module 350 predefines the event IDs of those events that it accept. The event IDs are hard-coded and unique for each server 180. Uniqueness of an event 700 within a server 180 is established by the combination of the source subsystem UID 722 and the event ID 712, and are used in combination to map events to handler routines as described above. Identifiers of the source host and the destination host may be passed as parameters in the SAL commands used to issue events 700. In such cases, the source host identifier and the source subsystem UID 722 together uniquely identify the sender (i.e., the source server and subsystem) of the event 700. The destination host identifier and the destination subsystem UID 724 uniquely identify the subsystem or system service module 350 targeted to receive the event. In one embodiment, the highest order bit of the event UID 712 is a "request bit" and indicates to the receiving subsystem how to map the event to the proper handler routine. All subsystems can optionally choose to handle events of another subsystem through such mechanisms as subscriptions. The event handler routines are mapped according to subsystem UID and event UID 712. Because the event being processed can either be directed or subscribed to, the request bit indicates whether to use the source or destination subsystem UID to map the event to the proper handler routine. The event header version identifier 714 defines the layout of the event header 710, such as the size and order of fields in the header 710. The event data version identifier 1416 implicitly defines the layout of the event data 730 included in the event 700. The event data offset identifier 720 indicates the offset from the event header 710 at which the event data 730 begins. The event data offset 720 equals the size of the event header 710. The channel identifier 726 is used to match a reply event to a request event. 5.1 Event Types Events can be one of several types including directed and notification events. 5.1.1 Directed Events Directed events are events that have a specified destination subsystem 300 when sent to the event delivery object 312. The specified destination includes a unique identification (UID) of the destination subsystem 300 and an identifier of the server 180 hosting the destination subsystem. Examples of directed events include notification events and the request and reply events described below. 5.1.1.1 Request-and-Reply Events Request events are subsystem specific directed events that send a request for service or functionality to another subsystem on the same server 180 or to a remote server in the server farm 110. Such request events contain codes that the destination subsystem can map onto known interfaces (i.e., event handler routines) to provide that service or functionality. Each request event includes a unique channel ID for use by the destination subsystem when creating a corresponding reply event. Reply events occur in response to request events. Each reply event is delivered as a directed event to the subsystem from which the corresponding request event originated. The reply event specifies the same channel ID and the same event buffer 380 used by the corresponding request event. The subsystem that sent the request event waits for the reply event from the event delivery object 312. The same channel ID indicates to the event delivery object 312 that the reply event is to pass directly to the destination subsystem rather than be placed in an event queue associated with the destination subsystem. The following pseudo-code embodies an example of a reply event handler routine that is called in response to receiving a request event. In particular, for the following example, the destination subsystem has a event handler routine, called OnGetSampleData(EventBuffer* pEvent), that is called in response to a GetSampleData request event. This event handler routine places data in the reply event buffer, pointed to by the pointer "pReplyEvent". RESULT Sample::OnGetSampleData(EventBuffer* pEvent) { if (SUCCESS==Create Reply Event(&pReplyEvent, SetSampleDataReply, event version, subsystem, size)) { put_data_in_event_buffer; res=PostEvent(pReplyEvent); // send event to the Event bus } delete(pEvent); // return res; } The OnGetSampleData reply event handler routine calls a CreateReplyEvent that creates a reply event to the original request event. As noted above, the reply event is placed in the event buffer used to hold the original request event (i.e., pointed to by pEvent), thus overwriting the request event. A new pointer, pReplyEvent, points to the reply event in the event buffer, and the old pointer, pEvent, is deleted. The Create_Reply_Event, as the name suggests, creates the reply event according to supplied input parameters. One input parameter is the identification of the reply event, here SetSampleDataReply, and the version of the reply event, here 1. All events are associated with an event ID 712, which together with the subsystem ID 722 of the source subsystem produces a unique identifier for that event. Another feature of the Create_Reply_Event is that this function automatically specifies the destination subsystem of the reply event, namely, the subsystem that originated the request event. The PostEvent command is one of the functions provided by the event bus API 392 for communicating with the event bus 310. Because the Create_Reply_Event function sets the destination subsystem of the event, the PostEvent command indicates where to the deliver the reply event (i.e., using the dispatch table). 5.1.1.2 Notification Event A notification event is an event that is directed to the subscription manager 362. Such event is dropped (i.e., ignored) by the subscription manager 362 unless there is an entry in the local subscription table 450 or remote subscription table 418 indicating that at least one subsystem 300 is interested in being notified of the occurrence of that event. Each subsystem keeps a list of events that can be subscribed to by other subsystems, and accordingly produces a notification event after issuing one of these potentially subscribed to events. 5.2 Event Delivery Commands In general, each subsystem 300 issues five types of commands to deliver events to the event bus 310: PostNotificationEvent, PostEvent, SendEventAndWait, Subscribe, and Unsubscribe. In brief overview, a PostNotificationEvent command sends a directed event to the subscription manager 362 as mentioned above. A PostEvent command sends a directed event to a destination subsystem and allows the source subsystem to immediately continue processing other tasks (that is, the PostEvent command immediately "returns"). A SendEventAndWait command sends a directed event to a destination subsystem and waits for a response causing the source subsystem to block until the response is received. A Subscribe command sends a notification event to register a subscription with the local subscription table 450 and/or remote subscription table 418. An Unsubscribe command sends a notification event to remove a previously registered subscription from the local subscription table 450 and/or the remote subscription table 418. 6.0 Basic Examples Referring back to FIG. 3, the following examples use one particular embodiment to illustrate the principles of the subject matter described above and are not intended to limit the subject matter of the invention in any way whatsoever. 6.1 PostEvent Command Referring also to FIG. 8A, when a source subsystem 300 seeks to communicate with a destination subsystem 300' on the same or different server, one method of communicating is for the source subsystem 300 to issue a PostEvent command to the event delivery object 312 through the event bus API 392. The source subsystem 300 determines (step 800) whether the identity of a target server hosting the destination subsystem 300' is needed. For example, a subsystem 300 preparing to issue an event to a peer subsystem 300 on another server 180 would need to determine the identity of the target server 180 hosting the peer subsystem. If the identity of a target server is needed, the source subsystem 300 communicates (step 802) with the service locator 354. In one embodiment, such communication occurs as a directed event to the service locator 354 delivered over the event bus 310. The directed event may request the identity of the target server or request that the service locator 354 forward the event 700 to the destination subsystem 300' on the target server. In the latter case, the event received by the service locator 354 from the source subsystem contains the event 700 that is to be forwarded. The service locator 354 delivers this contained event 700 to the event delivery object 312 with the target server specified as one of the parameters. In the embodiment shown in FIG. 8A, the source subsystem 300 does not deliver an event, but calls a function of the internal API 302 of the service locator 354. The service locator 354 then determines (step 804) the target server. In one embodiment, the service locator 354 returns (step 806) the identity of the target server to the source subsystem 300 so that the source subsystem 300 can issue (step 810) the PostEvent command to send the event 700 to the destination subsystem 300'. Alternatively, the service locator 354 issues (step 808) the PostEvent command to send the event 700 to the destination subsystem 300' on behalf of the source subsystem 300 over the event bus 310. For this case, the internal API 302 call contains the event 700 that is to be forwarded to the destination subsystem 300' on the target server. Upon receiving the event 700, the event bus 310 determines (step 812) whether the event 700 is local or remote from any destination host parameter included in the PostEvent command. If the destination subsystem 300' is remote, the event is delivered (step 814) to the transport layer 318 of the event bus 310 for subsequent transmission to the remote server 180' hosting the destination subsystem 300'. The transport layer 318 then transmits (step 816) the event 700 over the network connection 200 to the transport layer 318' on the remote server 180'. Operation of the transport layers 318, 318' is described in more detail in section 7.2. If the destination subsystem 300' is local, the event delivery object 312 of the event bus 310 determines (step 818) the entry point associated with the destination subsystem 300' and determines (step 820) whether the destination subsystem 300' is a single-threaded or multi-threaded subsystem. To determine the entry point, the event delivery object 312 examines the dispatch table 316 using the destination subsystem UID 724 of the event 700 as an index into the table 316. In embodiments having event queues, the dispatch table 316 identifies the event queue associated with the destination subsystem 300'. In one embodiment, the event queue indicates whether the destination subsystem 300' is multi-threaded. If the event queue indicates that the destination subsystem 300' is multi-threaded, the event 700 is not queued. The event delivery object 312 calls (step 822) the DispatchEvent of the subsystem API 306 of the destination subsystem 300', which causes execution (step 824) of the appropriate handler routine of the destination subsystem 300' for responding to the event 700. In an alternative embodiment, a thread executed by the destination subsystem 300' retrieves the request event 700 from the event delivery objects 312'. If the event queue indicates that destination subsystem 300' is single-threaded, the event delivery object 312 places (step 826) the pointer to the event buffer 380 holding the event 700 in the event queue associated with the destination subsystem 300'. The event delivery object 312 then starts (step 828) a new thread of execution that signals the destination subsystem 300', using the DispatchEvent function of the subsystem API 306, and delivers the event 700 from the event queue to the destination subsystem 300'. This new thread executes (step 824) the handler routine appropriate for the event 700. In one embodiment, the event delivery object 312 dispatches the event 700 (using DispatchEvent) to the destination subsystem 300' without placing the event 700 in the event queue if the event queue is empty when the event delivery object 312 is about to place the event 700 in the event queue. Again, in an alternative embodiment, a thread executed by the destination subsystem 300' retrieves the event 700 from the event queue, rather than the event delivery object 312 pushing the event 700 to the destination subsystem 300'. In one embodiment, the dispatch table 316 indicates whether the destination subsystem 300' has multi-threading capability. If the dispatch table 316 indicates that the destination subsystem 300' is multi-threaded, the event delivery object 312' calls the DispatchEvent function of the subsystem API 306' of the destination subsystem 300' as described above. Using the dispatch table 316 to store information regarding multi-threaded capability of subsystem makes the use of an event queue for a multi-thread capable subsystem unnecessary. 6.2 SendEventandWait Command Referring to FIGS. 9A-9D, another method for the source subsystem 300 to communicate with the destination subsystem 300' is for the source subsystem 300 to issue a SendEventandWait command to the event delivery object 312 through the event bus API 392. To start the process, subsystem 300 issues (step 902) a request event 700 using the SendEventAndWait command of the SAL 304 of the destination subsystem 300'. This request event 700 uses a channel identification and specifies the destination subsystem 300' in the destination UID 724. Because the request event 700 is an event for which a response is subsequently expected, the source subsystem 300 blocks further execution of the thread that generated the request event 700 until the response from destination subsystem 300' is received. While this thread is blocked, the source subsystem 300 can communicate with other subsystems through other threads. In this example, that source subsystem 300 seeks (step 904) a target server from the service locator 354. Note that not every event is sent to the service locator 354 for determining a target server; for some events, such as reply events, the source subsystem 300 does not need to use the service locator 354 because the target server is determined from the request event 700. As described above, the service locator 354 determines (step 906) the target server and returns (step 908) the identity of the target server to the source subsystem 300, and the source subsystem 300 sends the request event 700 to the event bus 310. Alternatively, the service locator 354 issues (step 910') the request event 700 to the event bus 310 on the source subsystem's 300 behalf. The specific action taken by the service locator 354 depends upon the actual request from the source subsystem 300. The request event 700 passes to the event delivery object 312 of the event bus 310. Assume that the service locator 354 determines the target server to be the remote server 180'. The event delivery object 312 then determines (step 912) from the destination host parameter of the SendEventandWait command that the destination subsystem 300' is on the remote server 180'. Because the destination subsystem 300' is remote to the source subsystem 300, the request event 700 passes (step 914) to the transport layer 318 on the server 180. The transport layer 318 then transmits (step 916) the request event over the network connection 200 to the transport layer 318' on the server 180'. The transport layer 318' passes (step 918) the request event 700 to the event delivery object 312' of the event bus 310'. The event delivery object 312' of the event bus 310' then determines (step 920) the entry point associated with the destination subsystem 300' and determines (step 922) whether the destination subsystem 300' is a single-threaded or multi-threaded subsystem as described above. If the destination subsystem 300' is multi-threaded, the request event 700 is not queued. The event delivery object 312' calls (step 924) the DispatchEvent of the subsystem API 306 of the destination subsystem 300', which causes execution (step 926) of the appropriate handler routine of the destination subsystem 300' for responding to the request event 700. If the destination subsystem 300' is single-threaded, the event delivery object 312' places (step 928) the pointer to the event buffer 380 holding the request event 700 in the event queue associated with the destination subsystem 300'. The event delivery object 312 then starts (step 930) a new thread of execution that signals the destination subsystem 300', using the DispatchEvent function of the subsystem API 306, and delivers (step 932) the request event 700 from the event queue to the destination subsystem 300'. This new thread executes (step 926) the handler routine appropriate for the request event 700. In one embodiment, the event delivery object 312' dispatches the request event 700 to the destination subsystem 300', bypassing the event queue if the event queue is empty when the event delivery object 312' is about to place the request event 700 in the event queue. The handler routine produces (step 934) a reply event 700' that is posted (step 936) by the destination subsystem 300' to the event delivery object 312' of the event bus 310'. The reply event 700' uses the same channel identifier provided by the source subsystem 300 when it issued the request event 700. After determining that the reply event 700' is for a remote server (here server 180), the event delivery object 312' then passes (step 938) the reply event 700' to the transport layer 318' on the server 180'. The transport layer 318' transmits (step 940) the reply event 700' to the transport layer 318 on the server 180 over the network connection 200. The event delivery object 312 of the event bus 310 receives (step 942) the reply event 700' through the transport layer 318 of the server 180 and delivers (step 944) the reply event 700' to the waiting thread (i.e., the thread that produced the request event 700). Because the reply event 700' uses the same channel identification used by the source subsystem 300 to initially issue the request event 700, the reply event 700' returns to the waiting thread (i.e., the waiting thread unblocks), bypassing the event queue (if any) associated with the source subsystem 300. If the reply event 700' does not return within a specified timeout period specified in the command, the waiting thread is released. The event delivery object 312 ignores the reply if the reply event 700' arrives after the timeout period expires. The source subsystem 300 executes the appropriate handler routine for the reply event 700'. In an alternative embodiment, a thread executed by the destination subsystem 300' retrieves the request event 700 from the event delivery object 312', and a thread executed by the source subsystem 300 retrieves the reply event 700' from the event delivery object 312. Thus, in this embodiment, the subsystems 300, 300' "pull" the event 700' in contrast to the above described embodiments in which the respective event delivery objects 312', 312 "push" the request event and reply events 700, 700' to the destination subsystems 300' and source subsystem 300, respectively. 6.3 Managing Dynamic Data Referring to FIG. 10, when a subsystem 300 of a server 180 needs to store or retrieve collector point data stored in dynamic store 240, that subsystem 300 transmits an event to the dynamic store system service module 356 resident on the server 180 (step 1002). The dynamic store system service module 356 determines if it knows which server 180 in the server farm 180' is the collector point of the record type sought by the subsystem 300 (step 1004). For example, the dynamic store system service module 356 may cache associations between the record type and the collector point, and access this cache upon receiving the event from the subsystem 300. If the dynamic store system service module 356 can determine the server collecting records of the type identified in the event, the dynamic store system service module 356 sends an event to the server 180 responsible for collecting such records (step 1006). If unable to determine the collector point, the dynamic store system service module 356 sends an event to the zone manager 358 seeking the address of the server that collects that record type (step 1008). Upon receiving that event (step 1010), the zone manager 358 determines (step 1012) if it is the master zone manager 358 for the zone. If the zone manager 358 is the zone master, then the zone manager 358 transmits to the dynamic store system service module 356 the identification of the server responsible for collecting events of identified type (step 1014). If the zone manager 358 is not the master, then the zone manager 358 sends (step 1016) an event to the zone master, which is known as a result of the master election. The zone manager 358 receives the server identification of the zone master (step 1018) and transmits (step 1014) the server identification to the dynamic store system service module 356. Upon receipt of this server identification, the dynamic store system service module 356 accesses the dynamic store 240 according to the event initially received from the subsystem 300. In the event that the zone master does not respond after a predetermined number of requests are sent, the zone manager 358 initiates an election for a new zone master, as described above. 7.0 Subsystems Whenever a dynamic store table is opened by a server 180 for the first time, the dynamic store contacts the zone master to determine the table owner. A request to the zone master for a table owner always succeeds assuming the requested table name is valid. Even if the table is not known to the zone master, an owner will be designated for it at the time of the request. Any failure to determine the table owner (other than invalid table name) is catastrophic, and will result in an error being propagated back to the component that initiated the connect request. After the zone master has returned the identity of the server that owns the table in question, the requesting server must contact the owner. If the connection attempt fails after a predetermined number of attempts, the requesting server resets its state and requests the zone master to again identify the table owner. This should eventually result in a new table owner being designated. After the record table has been successfully opened by contacting the table owner, the communication between requesting server and owning server settles into a set of insert, delete, update, and query requests. If a failure occurs while attempting to perform one of these operations after a predetermined number of attempts, the requesting server will contact the zone master to request a new owner. This process is executed as above. If a new table owner is selected by the zone master, the requesting server will first update the new owner with all local records. Since the new owner will need some time to receive updates from the other hosts in the zone before it will properly be able to deal with the incoming request in some embodiments the requesting server will have to wait for some amount of time before submitting the request. As described above in section 3.0, each subsystem 300 includes a subsystem access layer (SAL) 304 that defines the application program interface (API) commands to which the subsystem 300 is capable of responding. When one subsystem 300 needs to use the functionality of another subsystem 300', that one subsystem 300 calls the appropriate API command provided by the SAL 304 of that other subsystem 300'. In one embodiment, each SAL 304 is implemented as an object class having data members and member functions. The member functions use the event as a parameter in a command. These command functions include a PostSALEvent function (equivalent to a PostEvent function) and a SendSALEvent function (equivalent to a SendEventAndWait function). The data members include (1) a reference to the subsystem that created the SAL, (2) identification of the subsystem calling the member function using the event as a parameter and (3) identification of the destination subsystem for the event. When the source subsystem 300 needs to use the functionality of another subsystem 300', the source subsystem 300 creates an instance of the SAL class for that other subsystem 300' and calls the member functions provided by that SAL instance. When called, a SAL member function moves the event into an event buffer 380 and posts the appropriate pointer to the event to the event delivery object 312. For example, the called SAL member function sets the "request bit" in the event ID 712 and issues a SendSALEvent call to post the event and wait for a reply event. As discussed previously, the SendSALEvent call creates a unique channel ID 726 with which the destination subsystem sends a reply for this event to the source subsystem. Upon receiving a reply on the specified channel, the SAL 380 of the source subsystem extracts the data from parameters in the reply event and returns to the blocked thread that called the SAL member function. If the source subsystem and the destination subsystem are the same type of subsystem, but reside on different hosts, the source subsystem does not need to use the SAL 304 of the receiving subsystem (e.g., the persistent store system service module 352 on one server 180 to the persistent store system service module 352' of another server 180'). In such instances, the source subsystem already knows the events to use to communicate with the destination subsystem without needing to reference the SAL of the destination subsystem. In these embodiments, the source subsystem may directly post an event to the event bus directed to its peer residing on another host. 7.1 Transport Layer The transport layer 318 serves as the mechanism that allows subsystems 300 on different servers 180 to communicate with each other. The transport layer 318 corresponds to the Open Systems Interconnection (OSI) session and presentation layers in that it sends and receives event messages 700 via the server's network interface, performs encryption/decryption and compression, and manages connections. Connection management involves forming a connection to other servers in its server farm when there is an event message 700 to transmit and dropping the connection after a period of inactivity. The transport layer 318 handles two types of messages--control and event messages. Control messages are used by the transport layer 318 to determine the compatibility of encryption and compression capabilities (i.e., filters) for servers 180 on each side of the connection. In addition to resolving the transport capabilities of the receiving server during the negotiation cycle and establishing a connection, the control m | ||||||