Browse Prior Art Database

Technique for Dynamic determination of Monitoring Alert Thresholds for knowing the health of Performance metrics Disclosure Number: IPCOM000238447D
Publication Date: 2014-Aug-27
Document File: 4 page(s) / 87K

Publishing Venue

The Prior Art Database


Disclosed is a technique of dynamically deriving the optimal threshold values for system’s resources during runtime in JavaEE enterprise application servers. The method depends on Factor Dynamic Weight Determination System, which is explained in detail in the document. This system based on Factor Weight Determination, renders itself to continuous refinement of equations at runtime and continuous re-computation of weights depending on system state. This makes the overall system a Proactive Auto-tuning System. The disclosed technicque also includes an approach for Residual Analysis to Determine Relevance of Factors dynamically, which eliminates irrelevant factors which donot have a bearing on the system.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 01 of 4

Technique for Dynamic determination of Monitoring Alert Thresholds for knowing the health of Performance metrics

Table of Contents :

Detailed description

1. Factor Weight Determination System

2. Residual Analysis of Factors

3. Proactive Auto-tuning

Objective :

- Our objective is to dynamically determine the threshold values for server's resources, and thereby autotune the system holistically, in a JavaEE context.

- As an illustration, we will see how changes to configured number of Web container threads (WCT), or other factors, weighs on CPU usage, Memory usage, and request Throughput, and thereby be able to calculate the threshold value for WCT.

Weight Determination for Factors of influence :

- Steps #1 and #2 below outline how we determine how CPU usage, memory usage, and throughput are affected by WCT (and other factors) - and thereby obtain the weights of these factors.

- By doing so, we can come up with equations that describe these relationships which help us determine the thresholds and guide administrators to choose appropriate settings.

Extending the rationale to other factors of influence : This does not just apply to WCT. The same approach could be used for other factors that influence the CPU, Mem, and Tput levels. Any other Independent factors can be addressed by use of common multiple linear regression solution techniques to compute constant and weight values, and thereby the thresholds.

Detailed Description :

1. Factor Weight Determination System
- The Workings Outlined

Step #1

Objective : Determine the weight of WCT relative to CPU, Memory, and Throughput by varying WCT.

Approach :

- Vary WCT and record CPU, Mem, Tput levels

- Compute weights/constants to satisfy the following equations using common


Page 02 of 4

linear regression techniques (using ONLY data points before throughput reaches


(as illustrated in the graph of the figure below )

• CPU = c1 + W1 * WCT • Mem = c2 + W2 * WCT
• Tput = c3 + W3 * WCT

Where, Mem = Memory, Tput = throughput, WCT = number of 'Web container threads' configured
c1 = constant1 (the constant value 'c' in the equation y = c+mx)

W1 = Weight1 (the slope 'm' in the equation y = c+mx) c2 = constant2
W2 = Weight2
c3 = constant3
W3 = Weight3

Outcome of Step #1 : The relevant weights, constants and thereby Equations describing relationship between WCT and CPU, Mem, Tput

Step #2

Objective : Determine thresholds and provide guidance to administrator based on equations from Step#1

- Provide recommendation to administrator on the value of WCT that would result in CPU reaching 90% of maximum (or some other desired max threshold)

- (single factor example): 90 = c1 + W1 * WCT

- (multiple factor example): 90 = c1 + (W1 * WCT) + (Wx1 * FactorX) + (Wy1 * FactorY) + ...


Page 03 of 4

(Where Wx1 = weight of FactorX wrt CPU as the resource, Wy1 = weight of FactorY wrt CPU as the resource)

- Provide recommendation to admin...