Citation Context

...ing the information from deployed applications. Hauswirth and Chilimbi [14] developed an approach for memory leak detection in the field based on lowoverhead dynamic instrumentation. Nagpurkar et al. =-=[21]-=- focus on detecting performance problems in the field. However, they focus on embedded applications, rely on hardware profiling support, and identify hot code instead of long latencies. Adamoli and Ha...

Citation Context

... environment (JVM or CLR) [2, 9, 5, 17, 11, 21], across the system but agnostic to JVM-process internals [19, 15], employ complex software instrumentation [13, 23, 26], or require hardware extensions =-=[16, 18, 10, 4]-=-. To achieve efficient, full-system profiling currently requires multiple, complex, profiling tools and tedious, inaccurate, offline integration of profile data. To address these limitations, we inves...

Citation Context

...1] in virtualization contexts, and [6], [22], [23] in native contexts is one of the original objectives of VrtProf.Additional profiling tools also aid in this and other objectives; for example [24], =-=[25]-=-, [4], [5] are used for profiling across the kernel boundary, profiling embedded devices, managing hardware performance counters, and identifying and visualizing performance bottlenecks respectively. ...

Citation Context

...l to middlewares building and the notion of Java instance per Java application was rejected. Main reasons to this are related to language features. The problem with the new approach is the introduction of many QoS risks and even security risks. Isolation is a major concern in Java middlewares but addressing isolation at resources usage level is still a challenge. Resources accounting at virtual machine level is highly dependent on virtual machine implementation. For instances, thread’s model affects the way CPU usage is accounting. In a virtual machine like Kaffe you can apply sampling method [42] but it is a better and even cheaper solution to perform direct accounting. The same problem emerge in memory accounting, memory-management policy you use will affect the accounting subsystem. Even more, the target operating system affect the resources accounting subsystem. For instance, in Unix family the /proc pseudo-filesystem provide information about resource usage but the exact location and the amount of information vary among implementations. This particular issue is really important for applications’ developers because not all accounting systems offer the same accuracy. There is not ub...

Citation Context

... involved. Although there are some studies have been done on characterizing phase behavior for embedded systems, they still did their experiments by using SPEC CPU benchmarks such as SPEC CPU 2000 [4]=-=[5]-=-. A program phase is a set of intervals within the program’s execution that have similar behavior (e.g. in terms of required performance) regardless of temporal adjacency. The phase behavior provides ...

Citation Context

...s have been made for improving the performance counter infrastructure [16], attribution of performance-related events to particular instructions [5], and for sampling and processing of profiling data =-=[2, 13, 14, 19]-=-. Our cache miss classification mechanisms are synergistic with improvements in performance counters, sampling, and profiling infrastructure. Our mechanisms provide on-line identification of specific ...

Citation Context

...at the client remains small, while improving program performance. Although techniques similar to RCF’s constituent technologies have been previously considered for debugging [15], efficient profiling =-=[18, 33]-=-, and per-user online remote compilation of Java [14, 20, 32], RCF combines and extends these technologies into a single framework that targets, for the first time to our knowledge, low-footprint spec...