AU2002334163A1

AU2002334163A1 - State recovery

Info

Publication number: AU2002334163A1
Application number: AU2002334163A
Authority: AU
Inventors: William Box; Richard Hughes; Allan Jenkins; Ian Middleton; Keith Sterling
Original assignee: Jacobs Rimell Ltd
Current assignee: Jacobs Rimell Ltd
Priority date: 2001-10-15
Filing date: 2002-10-15
Publication date: 2003-07-03
Anticipated expiration: 2022-10-15

Description

State recovery

The invention relates to a method of recovering the modelled state of a system

In many electronic systems it is necessary to provide a framework to facilitate the storage of state information derived from monitoring entities within an external system. An example where this might be required is in a transaction based system with a high request inter-arrival rate such as that of monitoring a number of users logged on to a computer network.

In such systems, the problems of modelling the system and the state of the system are non-trivial. One approach that is used is to model the system using state counters associated with individual entities within the system, representing for example, the usage of the entity at any one instant in time. The required operations on these counters simply comprise increment and decrement requests.

Typically, in transaction based systems which have a high request inter-arrival rate, the option of storing this data in non-volatile memory can be dismissed due to the lengthy access times. Therefore volatile memory is the only option. However, in using volatile memory there is the risk that all state information will be lost in the case of a system failure of the monitoring system.

One known approach to the recovery of state information in a transaction based system, where one or more counters represent the correct, instantaneous state of the monitored system, when a failure of the monitoring application occurs and the state counters are lost, is to reset all the counters to zero, implicitly accepting that all the previous state data is lost and unrecoverable. The monitoring operation then continues as before and the counters are adjusted in response to incoming transaction requests.

The known approach suffers from the problem that the system will receive decrement requests which relate to the initial state before failure. However, the appropriate counters should not be decremented as they will tend to a negative value, thus misrepresenting the correct, instantaneous state of the monitored system. This occurs since there will be decrement requests received relating to increment requests received during the initial state. The problem with this is that the current state counters have no knowledge of what the state of the system was in the initial state. Therefore, any decrement requests received relating to increment requests received during the initial state will be deducted from the current state counters incorrectly.

The present invention seeks to provide a method of recovering the modelled state of a system.

According to the invention there is provided a method of recovering the state of a system, which system comprises at least one counter, which counter represents an instantaneous state of an element in the system, whereby the counter will increase in value in response to an increment request and decrease in value in response to a decrement request, wherein each increment request is paired with a decrement request.

The method of the invention advantageously pairs the increment and decrement requests. This permits the system in the recovery state to distinguish between known decrement requests, i.e. those corresponding to increment requests received in the recovery state, and unknown decrement requests, i.e. those corresponding to increment requests received in initial state. The unknown decrement requests can be discarded and hence have no effect on the recovery state. The counters will then tend to converge with the true state of the system.

In a preferred embodiment, the increment and decrement requests are paired using an index key shared between the increment and decrement requests. Preferably the system comprises a plurality of counters. Preferably the or each counter refers to a user session. Preferably, the state of the or each counter is held in volatile memory comprising a part of the system.

An exemplary embodiment will now be described in greater detail with reference to the drawing, in which:

Fig. 1 shows a graph of the state of a single counter A typical application where the monitoring of the state of a system is in the field of concurrency control such as a connection to an internet service provider or online licensing and gaming applications. Figure 1 shows a graph showing the state of a single counter over a period of time. The counter can refer to a user session such as a dialled up connection to an internet service provider. However, it is also possible to group individual users if this is appropriate so that a counter refers to all users in a particular geographic region or all users belonging to the same organisation. The former is applicable in the case of a vISP and the latter in the case of a corporate contract.

The period from t=0 to represents the initial state of the system, where ti represents the time at which the monitoring system fails. t₂ represents the time at which the system recovers and begins to process increment and decrement requests. t represents the time at which the recovery state represents the true state of the system being monitored, the time between t₂ and t₃ representing the recovery state. The thin line represents the true state of the system and the bold line represents the recovery state. The time t₃ will be reached once all of the decrement requests relating to increments from the initial state have been processed.

One method of pairing increment and decrement requests is to use a shared index key between increment and decrement requests associated with the usage of a particular entity, which can be defined in the data model of the system. Other methods of pairing increment and decrement requests are possible depending on the language and data constructs used in the implementation.

In addition to user sessions, it is possible for the counters to refer to other forms of user activity such as number of accesses to a piece of content or to a certain application or the volume of data transferred.

Claims

1. A method of recovering the state of a distributed networked system, which system comprises at least one counter, which counter represents an instantaneous state of an entity in a system, whereby the counter will increase in value in response to an increment request and decrease in value in response to a decrement request, wherein each increment request is paired with a decrement request.

2. A method according to Claim 1, wherein the increment and decrement requests are paired using an index key shared between the increment and decrement requests.

3. A method according to Claim 1 or Claim 2, wherein the system comprises a plurality of counters.

4. A method of recovering the state of a system according to any one of Claims 1 to 3, wherein the or each counter refers to a user session.

5. A method of recovering the state of a system according to any one of Claims 1 to 4, wherein the state of the or each counter is held in volatile memory comprising a part of the system.