| |
| software rejuvenation has been used to avoidsoftware (running image) aging, which is thephenomena of the gradual performance degradationof running software images due to unreleasedresources, accumulation of numerical errors, andfile system degradation. At the same time,software rejuvenation also can incur an overheadwhich should be balanced against the lossincurred due to unexpected outage caused by afailure. Thus, effective rejuvenation depends onthe right time. The time to rejuvenate is usuallyestimated based on the collection and statisticalanalysis of the values of aging related systemparameters. However, to our knowledge, there isno work on systematically identifying the agingrelated parameters. Thus, in this paper, wepropose a systematic method for identifying theseparameters, which is verified in the context ofXen. We firstly identify 2101 potential agingrelated system parameters that cover all the aspects of system, includingCPU scheduling, networking, memory/diskaccessing, virtual machine monitor communication and system kernel information. Then, we collect the values of theseparameters in both aging and healthy runningprocesses. After that, statistical test andmanual identification technique are used toobtain aging related parameters. We also verifiedthese parameters by a machine learning method,the results show that the parameters caneffectively represent the system performance andfurther system aging state. |
|
| Keywords:Software Engineering; Software Aging; Software Rejuvenation; Software Aging Parameter; Xen |
|
