X. Zhang, Z. Wang, N. Gloy, J. Chen, and M. Smith. System support for automated profiling and optimization. In ACM SOSP, October 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....that use transparent profiling to improve performance of future executions, although none of these systems perform compilation optimization at runtime. DCPI [6] performs sampling of the program counter (PC) using interrupts generated by the performance counters on the ALPHA processor. Morph [82] describes a similar system in which timer interrupts are used to trigger PC sampling. Digital FX 32 [52] is a system that allows x86 Win32 applications to execute transparently on Windows NT Alpha systems. Digital FX 32 profiles x86 applications during emulation, and later translates them into ....

Xiaolan Zhang, Zheng Wang, Nicholas Gloy, J. Bradley Chen, and Michael D. Smith. System support for automated profiling and optimization. In Proceedings of the 16th Symposium on Operating Systems Principles (SOSP-97), Operating Systems Review, 31(5), pages 15--26, October 5--8 1997.

....and call statistics for functions (gprof ) or basic blocks (pixie) can be recorded. 1 The second category of profiling tool includes those based on statistical sampling of application execution. Notable here are prof [15] which has been in use for a very long time, and several new tools, Morph [16] and the Digital Continuous Profiling Infrastructure (DCPI) 1] Prof uses program counter sampling to generate an estimate of the execution time and call statistics for the functions in the application. Morph and DCPI operate with sufficiently low overhead that they may be run continuously. DCPI ....

Xiolang Zhang, Zheng Wang, Nicholas Gloy, J. Bradley Chen, and Michael D. Smith. System support for automated profiling and optimization. In Proceedings of the Sixteenth ACM Symposium on Operating System Principles, pages 15-- 26. ACM SIGOPS, October 1997.

....any modification to kernel source code. Although the System Monitor currently operates only on the Linux operating system, this design approach should enable it to be relatively portable to other operating systems. The design of the System Monitor is similar to the sampling components of Morph [100] and DCPI [3] The present implementation samples system activity when triggered by the digital multimeter. Each twelve byte sample records the value of the program counter (PC) and the process identifier (PID) of the currently executing process, as well as additional information such as whether ....

....Spectra. 8.1 Energy measurement While PowerScope is the first energy profiler, its development was aided by previous work in CPU profiling. In particular, the implementation of the System Monitor is closely related to similar sampling based profilers designed to run continuously, such as Morph [100] and DCPI [3] 127 128 CHAPTER 8. RELATED WORK Other than the profiling technique used by PowerScope, there are two main approaches to measuring application energy usage. The first approach, which I shall refer to as energy accounting, measures the energy consumption of specific system ....

Zhang, X., Wang, Z., Gloy, N., Chen, J. B., and Smith, M. D. System support for automated profiling and optimization. In Proceedings of the 16th ACM Symposium on Operating Systems Principles (SOSP), Saint-Malo, France, October 1997.

....a particular path through a program, possibly penalizing other paths. For example, procedures can be inlined based on their execution frequency. Many commercial compilers exploit profiling data during recompilation. Morph, a research system, can profile and optimize binaries without source [71]. This is only beneficial, however, when the resulting executable maintains the same behavior exhibited by the program during its profiled run. In runs manifesting different behavior, feedbackdirected optimization can hinder performance. Massalin s Synthesis kernel and the continuing Synthetix ....

Catherine Xiaolan Zhang, Zheng Wang, Nicholas C. Gloy, J. Bradley Chen, and Michael D. Smith. System support for automated profiling and optimization. In Symposium on Operating Systems Principles, pages 15--26, 1997. 23

....an application as feedback into a compiler to make better optimization decisions for future executions. In addition, there are fully automated profiling systems that use transparent low overhead profiling to improve performance of future executions. Such systems include Digital FX 32 [33] Morph [48], and DCPI [4] In contrast, this paper focuses on using profiling information of an application s execution to help optimize that same execution of the application. Nevertheless, our architecture can be extended to handle offline profile directed compilation schemes; for example, the controller ....

X. Zhang, Z. Wang, N. Gloy, J. B. Chen, and M. D. Smith. System support for automated profiling and optimization. In Proceedings of the 16th Symposium on Operating Systems Principles (SOSP-97), Operating Systems Review, 31(5), pages 15--26, Oct. 5--8 1997.

....of the target function. To finish, Detours restores the original page permissions on both the target and trampoline functions and flushes the CPU instruction cache with a call to FlushInstructionCache. 2.2. Payloads and DLL Import Editing While a number of tools exist for editing binary files [10, 12, 13, 17], most systems research doesn t require such heavy handed access to binary files. Instead, it is often sufficient to add an extra DLL or data segment to an application or system binary file. In addition to detour functions, the Detours library also contains fully reversible support for attaching ....

....but can invoke its functionality at any point through the trampoline. Our unique trampoline design makes it trivial to extend the functionality of existing binary functions. Recent research has produced a class of detailed binary rewriting tools including Atom [13] Etch [12] EEL [10] and Morph [17]. In general, these tools take as input an application binary and an instrumentation script. The instrumentation script passes over the binary inserting code between instructions, basic blocks, or functions. The output of the script is a new, instrumented binary. In a departure for earlier ....

Zhang, Xiaolan, Zheng Wang, Nicholas Gloy, J. Bradley Chen, and Michael D. Smith. System Support for Automated Profiling and Optimization. Proceedings of the Sixteenth ACM Symposium on Operating System Principles. Saint-Malo, France, October 1997.

....early work such as the Smalltalk 80 [17] and Self 91 [12] systems, as well many of today s JIT Java compilers [1, 24, 33] There are also fully automated profiling systems that use transparent profiling to improve performance of future executions. Such systems include Digital FX 32 [21] Morph [34], and DCPI [4] However, these systems do not perform compilation optimization at runtime. 12 7 Conclusions This paper describes an empirical study of selective optimization using the Jalape no JVM. Two categories of benchmarks, short running (1.1 4.8 seconds) and longer running (16.6 70.6 ....

Xiaolan Zhang, Zheng Wang, Nicholas Gloy, J. Bradley Chen, and Michael D. Smith. System support for automated profiling and optimization. In Proceedings of the 16th Symposium on Operating Systems Principles (SOSP-97), Operating Systems Review, 31(5), pages 15--26, October 5--8 1997. 15

....system used in this work is also in this category, although work is currently underway to make it an adaptive system. There are also fully automated profiling systems that use transparent profiling to improve performance of future executions. Such systems include Digital FX 32 [19] and Morph [31]. However, these systems do not perform compilation optimization at runtime. 5 Conclusions This paper described an empirical study of selective opt compilation using the Jalape no JVM and the SPECjvm98 benchmarks. The main conclusions are ffl The Best number of methods to opt compile varies ....

Xiaolan Zhang, Zheng Wang, Nicholas Gloy, J. Bradley Chen, and Michael D. Smith. System support for automated profiling and optimization. In Proceedings of the 16th Symposium on Operating Systems Principles (SOSP-97), volume 31,5 of Operating Systems Review, pages 15--26, New York, October 5--8 1997. ACM Press. 13

....of profile collection and the running of the off line analysis tool. 3.2. The System Monitor The System Monitor consists of a user level daemon process and a small set of modifications to the NetBSD kernel. Its design is similar to the sampling components of continuous profilers such as Morph [9] and DCPI [1] Our current implementation samples system activity when triggered by the digital multimeter. Each twelve byte sample records the value of the program counter (PC) and the process identifier (PID) of the currently executing process, as well as additional information such as whether ....

....is the first tool that maps energy consumption to program structure. At the same time, our development of the tool was motivated by previous work in CPU profiling. In particular, the implementation of the System Monitor is closely related to similar components of continuous profilers such as Morph [9] and DCPI [1] Several researchers have previously investigated power management in the context of wireless networks and disks. Stemm and Katz [7] measured the power consumption of several network interfaces. They also presented simulation results showing how powering down the network interface ....

X. Zhang, Z. Wang, N. Gloy, J. B. Chen, and M. D. Smith. System Support for Automated Profiling and Optimization. In Proceedings of the 16th ACM Symposium on Operating Systems and Principles, Saint-Malo, France, October 1997. 9

....of a page fault is so great, that such costs can in fact be justified. 2.2 Profiling There has been a renewed interest in profiling as means for smart optimization. The DCPI system [1] provides support infrastructure for sample based CPU profiling to sub instruction accuracy. The Morph system [5] builds on top of DCPI to gather profiling information on executables and then rewrite them for better CPU performance. 2.3 Modern systems Let us take the BSD 4.4 Unix [3] as a case study since it incorporates in a production system most of the leading edge OS research. The BSD 4.4 virtual ....

Xiaolan Zhang, Zheng Wang, Nicholas Gloy, J. Bradley Chen, and Michael D. Smith. System support for automated profiling and optimizatio. SOSP, 1997.

....of the target function. To finish, Detours restores the original page permissions on both the target and trampoline functions and flushes the CPU instruction cache with a call to FlushInstructionCache. 2.2. Payloads and DLL Import Editing While a number of tools exist for editing binary files [8, 10, 11, 15], most systems research doesn t require such heavy handed access to binary files. Instead, it is often sufficient to add an extra DLL or data segment to an application or system binary file. In addition to detour functions, the Detours library also contains fully reversible support for attaching ....

....but can invoke its functionality at any point through the trampoline. Our unique trampoline design makes it trivial to extend the functionality of existing binary functions. Recent research has produced a class of static binary rewriting tools including Atom [11] Etch [10] EEL [8] and Morph [15]. In general, these tools take as input an application binary and an instrumentation script. The instrumentation script passes over the binary inserting code between instructions, basic blocks, or functions. The output of the script is a new, instrumented binary. In general, these static binary ....

Zhang, Xiaolan, Zheng Wang, Nicholas Gloy, J. Bradley Chen, and Michael D. Smith. System Support for Automated Profiling and Optimization. Proceedings of the Sixteenth ACM Symposium on Operating System Principles. Saint-Malo, France, October 1997.

....run the program with a fixed or arbitrary training workload and use the profile to guide optimizations. Based on the assumption that users are similar, the training workload is considered to be representative. Spike [CG97] is an example of this approach. Some recent systems, such as Morph [ZW97] and FX 32 [HH97] extend the optimization process beyond an application s release by profiling and optimizing the application continuously while it is used . Based on the assumption that users are different, the current versions of Morph and FX 32 operate on a per user basis. Another ....

X. Zhang, Z. Wang, N. Gloy, J. B. Chen, and M. D. Smith, "System Support for Automated Profiling and Optimization." In Proceedings of the 16th ACM Symposium of Operating Systems Principles, ACM, pages 15-26, October 1997.

....Spike looks for exact matches based on each procedure s control flow structure, and propagates the entire profile for each unchanged procedure. Our binary matching and profile propagation occur on a finer level, and we seek to re use profile information even for procedures that have changed. Morph [ZWG97] performs profile re use in the context of program re optimization. When a program module is optimized, the optimizer produces a basic block mapping between the intermediate representations of the old and new versions so that existing profiles can be converted for re use. Morph s special ....

X. Zhang, Z. Wang, N. Gloy, J. B. Chen, and M.D. Smith, "System Support for Automated Profiling and Optimization." Proc. of the 16th ACM Symposium of Operating Systems Principles, October 1997. Appendix: An example of control tree matching

No context found.

X. Zhang, Z. Wang, N. Gloy, J. Chen, and M. Smith. System support for automated profiling and optimization. In ACM SOSP, October 1997.

No context found.

X. Zhang, Z. Wang, N. Gloy, J. Chen, and M. Smith. System support for automated profiling and optimization. In Proc. of Symposium on Operating Systems Principles, Oct. 1997.

No context found.

C. X. Zhang, Z. Wang, N. C. Gloy, J. B. Chen, and M. D. Smith. System support for automated profiling and optimization. In Symposium on Operating Systems Principles, pages 15--26, 1997.

No context found.

Catherine Xiaolan Zhang, Zheng Wang, Nicholas C. Gloy, J. Bradley Chen, and Michael D. Smith. System Support for Automated Profiling and Optimization. In Proc. 16th Symposium on Operating System Principles, pages 15--26, October 1997.

No context found.

X. Zhang, Z. Wang, N. Gloy, J. B. Chen, and M. D. Smith. System Support for Automated Profiling and Optimization. In Proceedings of the 16th Symposium on Operating Systems Principles (SOSP-97), pages 15--26, October 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC