Browse Prior Art Database

Applying predictive technique for self-adaptive polling in WebSphere Adapters

IP.com Disclosure Number: IPCOM000186428D
Original Publication Date: 2009-Aug-20
Included in the Prior Art Database: 2009-Aug-20
Document File: 5 page(s) / 270K

Publishing Venue

IBM

Abstract

Disclosed is a method which can be used for dynamic polling in WebSphere* Adapters based on exponential weighted moving average. One of the modes of connectivity of WebSphere Adapters to Enterprise Information Systems is inbound where events generated in the EIS are polled by the adapter. The adapter uses two parameters to control polling behavior - Poll Frequency which is the time interval after which the adapter will poll for events from the EIS and Poll Quantity which is the maximum number of events which can be fetched by the adapter during a single polling cycle. In current implementations, the Poll Quantity and Poll Frequency cannot be changed once the adapter starts executing. The core idea of this invention is to allow adapters to support adaptive polling which involves adjusting the values of Poll Quantity and Poll Frequency based on the changes in the number of events generated in the EIS. Our invention proposes an algorithm based on exponential weighted moving averages. Adopting a dynamic self adaptive polling algorithm in WebSphere Adapters helps to vary the Poll Quantity proportionally with number of events in the EIS, therby reducing overheads and number of calls to the EIS.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 38% of the total text.

Page 1 of 5

Applying predictive technique for self-adaptive polling in WebSphere Adapters

Disclosed is a method which can be used for dynamic polling in WebSphere Adapters based on exponential weighted moving average.

WebSphere JCA Adapters provide bidirectional connectivity between J2EE/SCA applications and Enterprise Information Systems. They provide outbound connectivity wherein applications can send requests to the EIS and inbound connectivity where events generated in the EIS are

polled by the adapter.

The adapter uses two parameters to control polling behavior -

                                     oll Frequency which is the time interval after which the adapter will poll for events from the EIS and

Poll Quantity

                                                  which is the maximum number of events which can be fetched by the adapter during a single polling cycle.

In current implementations, the

P

                                oll Frequency cannot be changed once the adapter starts executing. The implication of this is that a maximum number of events equal to the

oll Quantity and

 oll Quantity are polled even if a larger number of events are generated at the EIS. This results in more calls to the EIS which increases overheads. However if the

P

                                        oll Quantity is dynamically adjusted in proportion to the number of events in the EIS, the number of calls made to the EIS will be appropriate.

One solution to this problem has been proposed in [ 1 ] which describes an Intelligent Polling Algorithm which dynamically computes

P

                                            oll Frequency using cumulative averages. There are drawbacks to the approach proposed by [ 1 ].

Firstly, the algorithm averages a set of observations of number of events polled over a window (which is a fixed time interval over which a series of values are considered).As a result, short term peaks or spikes which should be ideally ignored also get included in the computation of averages resulting in an inaccurate estimate for the poll quantity. Secondly, as the algorithm uses simple average, equal weightage is assigned to all the observations. However, less weightage should be assigned to older observations as compared to observations which are more recent which should contribute more towards the computed estimate of

.

oll Quantity . If this

                             oll Quantity may get evened out and if the window size is too small then short term peaks and fluctuations will not get ignored. As the algorithm uses a fixed window size, it is necessary that an optimum value of window be chosen. Another drawback of this algorithm is that as the value of

window is too large then trends in changes of

                                   oll Quantity changes only at the end of each window, the adapter may not always be able to react quickly to increases/decreases in the number of events generated in the EIS.

Due to the above mentioned drawbacks, there is a need for an additional solution which provides a better estimate of the

P

.

The core idea of this invention is to allow adapters to s...