Dynamic Tracking Performance Of The System Model And Fleet Monitoring Software Based On Dynamic Tracking Technology Design And Realization

In the last 10 years,personal computer becomes cheaper and more powerful.The trend is that more and more PC clusters are applied as servers because they are cheap and as powerful as high-end servers and commercial clusters.Most of them are built with open-source software and free software.This raises the problem that there are lack of useful management software and monitoring software on PC cluster even though the relevant software is well-designed on single PCs.It is because most of them are designed for general purpose usage and designed for universal clusters. Another problem is that the design of these cluster management software and cluster monitoring software are not extensible and tailorable.Hence,we want to design and implement a new cluster monitor which is "dynamic","extensible","tailorable".And it will be deployed on small-scale non-comercial PC clusters only.In the paper,we give out a survey about the clusters and the monitoring software first.After this,we put forward our criterions about what the cluster monitor should be.We build our cluster monitor,DDTC/DDTS,under the direction of these criterions.The data-scope problem and extensibility problem will be well-solved because DDTC/DDTS is based on Dynamic Tracing System,which could tracing system as you want.Through our plug-in mechanism,the monitor's function could be extended easily by developing new plug-ins.Another significant criterion is monitors' run-time affection on it resident systems.Before designing DDTC/DDTS,we model the performance of DTrace,and give expressions about the relation between probes and CPU usage.It is easy to estimate the affection brought in by DTrace scripts through this model.It is proved that the affection is less than one percent in general.