مقاله انگلیسی رایگان در مورد تکنیک های ثابت مهندسی در سیستم های محاسباتی ابر – IEEE 2017

DYNAMIC INVARIANTS are properties of a program or a system expected or observed to hold during executions. Dynamic program invariants can be inferred from execution traces as likely invariants [1], a relaxed form modeling properties which hold during one or more executions, though not necessarily over all possible executions. Program likely invariants have been shown to support several software engineering activities [2] [3] [4] [5]. Likely system invariants [6] are attractive for modeling runtime behavior of data centers and cloud-based utility computing systems from a service operation viewpoint. They are operational abstractions of their dynamics. Due to the size and complexity of such systems, it is very hard for human operators to detect application problems in real time. Especially transient or silent errors occur rarely – e.g. in case of overload, timing issues and exceptions – and often do not cause an immediately observable failure such as a crash or hang, hence are hard to detect. Typically, likely system invariants hold in normal operating conditions; as such, their violations are considered symptoms of execution malfunctions. By monitoring execution and checking for broken invariants, it is possible to automatically detect failures [7] and to request actions to the operations personnel (e.g. jobs re-execution). Defining invariants is pretty natural for cluster computing or Software-as-a-Service (SaaS) platforms, and generally for systems performing batch work, providing services to applications often consisting of jobs, in turn comprising tasks. These systems include monitoring and logging facilities1 collecting metrics – e.g., job/task completion time, resource usage and status codes – which can be used to establish invariants. Indeed, likely system invariants have been shown to be effective for modeling execution dynamics in a variety of service computing systems [8], and for supporting a range of operational activities, including capacity planning, detecting anomalous behaviors [9], silent failures [10], and violations of service level agreements [11].