ENDO, Y., WANG, Z., CHEN, J. B., AND SELTZER, M. Using latency to evaluate interactive system performance. In Proceedings of the 1996.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....future to investigate the operating system automatically monitoring the system and detecting when to initiate a dynamic switch, for example, when a file grows too large, when the contention for a shared lock increases, or when its reference pattern changes. There has been other work in this area [7, 27] and we would like to be able to leverage it to help K42 self adapt. We believe that dynamic on the fly customization will again result in the radical performance advantages that researchers demonstrated with customizable operating systems. ....

Y. Endo, Z. Wang, J. Chen, M. Seltzer. Using latency to evaluate interactive system performance. In Proc. 2nd Symp. on Operating Systems Design and Implementation, Seattle WA, October 1996. 76

....the kernel s clock resolution and improving the billing accuracy of time consumed by interactive processes allows the system to reduce the dispatch latency of modern interactive processes. This is significant because latency is an important metric for the servicing of interactive processes [5, 4]. We measured a process s dispatch latency for each event as the interval between when the process is inserted into the run queue as a result of the event, and the next time it was scheduled (the Quake application, which is always runnable, never experiences dispatch latencies, so it is not ....

Y. Endo, Z. Wang, J. B. Chen, and M. Seltzer, "Using latency to evaluate interactive system performance ". In 2nd Symp. Operating Systems Design & Implementation, pp. 185--199, Oct 1996.

.... deadline based scheduling, explicit CPU or resource management [Mercer et al. 94, Nieh Lam 97, Jones et al. 97, Banga et al. 99] priority inheritance [Sha et al. 90] finegranularity clock and timer services [Jones et al. 96] and bounded response time for essential system services [Mogul 92, Endo et al. 96] Features it does have include elevated fixed real time thread priorities, interrupt routines that typically re enable interrupts very quickly, and periodic callback routines. Under Windows NT not all CPU time is controlled by the scheduler. Of course, time spent handling interrupts is ....

Yasuhiro Endo, Zheng Wang, J. Bradley Chen, and Margo Seltzer. Using Latency to Evaluate Interactive System Performance. In Proceedings of the Second USENIX Symposium on Operating Systems Design and Implementation (OSDI '96), Seattle, pages 185-199, October 1996.

....to be studied. Chen, et al. compared the performance of different flavors of Windows and NetBSD on micro and throughput oriented application benchmarks using hardware cycle counters [1] They followed up this work with a latency oriented evaluation more suit8 able for interactive applications [9], and TIPME, a tool and framework for addressing latency problems in interactive applications [8] Bershad, et al. have characterized the resource usage patterns of desktop and commercial applications running on Windows by instrumenting applications [16, 17] Perl and Sites studied the ....

ENDO, Y., WANG, Z., CHEN, J. B., AND SELTZER, M. Using latency to evaluate interactive system performance. In Proceedings of the

....then the di erence between those two values indicates the number of cycles consumed by the old=read tsc( while (1) new=read tsc( if (new old THRESHOLD) break; old=new; Figure 4: Measuring the cost of an interrupt interrupt handler. A similar technique was used by Endo et al. [Endo96] to measure the time to process user generated events in an interactive system. The measurement system was a 200MHz Pentium Pro running RedHat Linux 5.1 with version 2.0.30 of the Linux kernel. The interrupt measured is an upComplete interrupt generated by a 3Com 3c905 Fast Ethernet PCI card ....

Y. Endo et al.; \Using Latency to Evaluate Interactive System Performance", 2nd Symposium on Operating Systems Design and Implementation (October 1996), pp.185-199.

....specifically at PDA class devices. Standard batch oriented benchmarks, such as compilation or simulation, are abandoned in favor of applications one might find on a PDA. We employ three benchmarks, each with a contrasting workload: Address Book (UI) User interfaces differ from data processing [5] in that they are characterized by an alternation between idle (waiting for the user) and maximum performance processing. The processing bursts are typically small, such as drawing an activated button, but are occasionally larger, such as a spreadsheet update. Ideally, the delay for these ....

....before its deadline, effectively completes at its deadline. We therefore define clipped delay with deadline to be: For human interactive tasks (UI) the deadline is dependent on human perception time. A lower bound of 10 ms given from the limitations of human visual [7] and audio percep a 1. [5] discusses a similar metric for the bechmarking of workstation user interfaces. Figure 1: Event Delay Impulse Graphs 0 20 40 60 80 100 120 140 160 Event Delay (ms) Event Number (404) 312) 637) UI Event Delays 1 855 1 1.5 2 2.5 3 3.5 Event Delay (ms) Event Number ....

Y. Endo, Z. Wang, J. B. Chen, and M. Seltzer, "Using Latency to Evaluate Interactive System Performance," Proc. 2nd Symp. on Operating Systems Design and Implementation, Nov. 1996.

....operating system code on caches, branch predictors and TLBs, and unveils characteristics of context switches in Windows NT. This work also allows for a comparison of real world applications against popular CPU intensive benchmarks. While there have been some studies on x86 in the past few years [2, 3, 4, 5], execution behavior of x86 applications including operating system activity and context switching has not been adequately studied. Architects and system designers have realized the importance of including operating system activity in traces and there has been significant progress in this ....

....are necessary. These results show that simulating in only one address space can produce optimistic TLB results for desktop and database applications. 5. RELATED WORK Recently there has been some progress in understanding applications running on the x86 platform. Chen et al.# [3] and Endo et al.# [5] studied different operating systems running on Intel x86 processors using performance monitoring counters. Lee et al.# [4] performed a study illustrating the execution Understanding the Impact of x86 NT Computing on Microarchitecture 223 0 10 20 30 40 50 60 70 80 90 100 db ....

Y. Endo, Z. Wang, J. B. Chen, and M. I. Seltzer, "Using latency to evaluate interactive system performance," in Proceedings of the 2nd USENIX Symposium on Operating Systems Design and Implementation (OSDI), pp. 185--199, Oct 1996.

....with this approach is that in interactive applications not all parts of the program s execution are equally important. The relevant metric for measuring the quality of performance is not end to end throughput, but response time: the time it takes the computer to respond to user initiated events [11]. This time has also been called wait time referring to the fact that during these episodes the user is actively waiting for the computer to complete the task. The beginning of an interactive episode is initiated by the user and is usually signified by a GUI event, such as pressing a mouse ....

....characterization and power management. We aim to provide a broad overview of these fields with an emphasis on papers that are directly relevant to our studies. 2. 1 Characterization of interactive applications Various papers have dealt with the characterization of desktop applications [27] 6][11]. In [11] Endo et al. performed a detailed analysis of interactive performance in a uniprocessor Windows NT environment. In their paper they point out the need to measure response time, instead of throughput for the entire benchmark, and explore the changes in response times on different ....

[Article contains additional citation context not shown here]

Y. Endo, Z. Wang, J. B. Chen, and M. I. Seltzer. Using Latency to Evaluate Interactive System Performance. 2nd Symposium on Operating Systems Design and Implementation, pp. 185-199, October 1996.

....It also makes comparison with other papers much harder. 2.2. Why existing benchmarks don t suffice Most existing benchmarks, while usually reproducible, are not realistic. I will discuss a few of the most widely used benchmarks, although space constraints lead me to oversimplify. Endo et al. [9] provide a similar discussion, with different emphasis. 2.2.1. Vendor benchmarks Vendor benchmarks that measure operating system performance include AIM (from AIM Technology) and SYSmark (from the Business Applications Performance Corp. and SPECweb96 and SFS97 (LADDIS) from the Standard ....

Y. Endo, Z. Wang, J. B. Chen, and M. Seltzer. Using Latency to Evaluate Interactive System Performance. In Proc. 2nd Symp. on Operating Systems Design and Implementation, pp. 185-199. USENIX, Seattle, WA, Oct., 1996.

....she reclaims her workstation. This delay is a result of the workstation s memory contents being paged out to accommodate guest data while the workstation was idle. Measuring this delay directly is difficult it poses some of the same challenges that arise in benchmarking interactive applications [10]. This delay depends on the size of the memory context that will need to be re established when the owner reclaims her workstations. We refer to this metric as the memory context size (MCS) and use it as a measure of the cost of exploiting idle memory. Note that the actual delay experienced by the ....

Y. Endo, Z. Wang, J. B. Chen, and M. Seltzer. Using latency to evaluate interactive system performance. In Proc. of OSDI'96, pages 185--200, 1996.

No context found.

ENDO, Y., WANG, Z., CHEN, J. B., AND SELTZER, M. Using latency to evaluate interactive system performance. In Proceedings of the 1996.

No context found.

Endo, Y., Wang, Z., Chen, B., Seltzer, M. "Using Latency to Evaluate Interactive System Performance". Proceedings of the 1996 Symposium on Operating System Design and Implementation(OSDI).

....as an application specific graft. 5.4 Interrupt Latency While the time spent waiting for an interrupt to take place is compulsory, the time between the occurrence of the interrupt and when the application runs is needless. Previous studies of latency point out that slow systems irritate users [7]. By reducing this latency we can improve not only overall application performance, but also overall user satisfaction. In order to measure this latency, we time stamp interrupts as they arrive, and compute the difference between this time and when the process to which the interrupt is delivered ....

Endo, Y., Wang, Z., Chen, J., Seltzer, M., "Using Latency to Evaluate Interactive System Performance, " Proceedings of the 2nd OSDI, Seattle WA, October 1996.

....tend to be used by most users. 1.2 Related Work Although the majority of today s personal computers run mostly interactive applications on Windows systems, there has been little research on how people use these programs. Several research projects investigated Windows operating system performance [CE96, EW96, PS96]. A recent paper [LC98] presented measurements and simulation results of instruction set and architectural characteristics during program execution on x86 Windows NT. These projects focused on the characteristics and comparison of the general system performance, while this paper focuses on the ....

Y. Endo, Z. Wang, J. B. Chen, and M. I. Seltzer, "Using Latency to Evaluate Interactive System Performance." In Proceedings of the Second Symposium on Operating Systems Design and Implementation, USENIX Association, pages 185-199, October 1996.

....milliseconds, not the microsecond scale that is often the target of detailed performance tuning. Users perceptions of performance are closely related to response time and the variability of response time, both of which can be quantified with moderate effort, using some newer tools and techniques [7][8] However, to the best of our knowledge, there are no tools available for interpreting a collection of event latencies and determining which ones actually irritate users. For example, if an event s latency is below the threshold of human perception, that latency contributes nothing to user ....

....of perceptibility, there are no guidelines by which to assess the impact of the delay, but the relationship between delay and irritation is practically guaranteed to be nonlinear. Moreover, previous studies have argued that user expectation is a critical component of user perceived performance [6][7]. There is a qualitative difference between a five second delay echoing a keystroke and a five second delay starting up an application. Unlike latency, expectation is difficult to quantify because of its psychological aspect and because it is partially a reflection of the performance ....

[Article contains additional citation context not shown here]

Yasuhiro Endo, Zheng Wang, J. Bradley Chen and Margo Seltzer, "Using Latency to Evaluate Interactive System Performance," Proceedings of the Second Symposium on Operating System Design and Implementation, October 20 1996, pages 185--199.

.... basing performance tuning decisions solely on the performance metric that represents only one of many aspects of user perceived performance, system designers are sometimes misled into making design and implementation decisions that have an overall negative impact on user perceived performance [3][11]. As a result, interactive systems are tuned suboptimally. Real performance problems performance problems that actually annoy users, such as unexpectedly long response times remain uncorrected. The most important performance metric of interactive systems is user perceived performance. ....

....of such a measurement methodology. 1. 1 User perceived performance In interactive systems, users perceptions of performance are closely related to response time and the variability of response time, both of which can be quantified with moderate effort, using some newer tools and techniques [11][15] However, to the best of our knowledge, there are no tools available for interpreting a collection of event latencies and determining which ones actually irritate users. For example, if an event s latency is below the threshold of human perception, that latency contributes nothing to user ....

[Article contains additional citation context not shown here]

Yasuhiro Endo, Zheng Wang, J. Bradley Chen and Margo Seltzer, "Using Latency to Evaluate Interactive System Performance," Proceedings of the Second Symposium on Operating System Design and Implementation, October 1996, pages 185--199.

No context found.

ENDO, Y., WANG, Z., CHEN, J. B., AND SELTZER, M. Using latency to evaluate interactive system performance. In Proceedings of the 1996.

No context found.

ENDO,Y.,WANG, Z., CHEN, J. B., AND SELTZER,M. Using latency to evaluate interactive system performance. In Proceedings of the 1996.

No context found.

Yasuhiro Endo, Zheng Wang, Bradley Chen, and Margo Seltzer. Using latency to evaluate interactive system performance. In Proceedings of the 1996 Symposium on Operating System Design and Implementation, Seattle, WA, October 1996. 158

No context found.

Yasuhiro Endo, Zheng Wang, Bradley Chen, and Margo Seltzer. Using latency to evaluate interactive system performance. In Proceedings of the 1996 Symposium on Operating System Design and Implementation, Seattle, WA, October 1996.

No context found.

ENDO,Y.,WANG, Z., CHEN, J. B., AND SELTZER,M. Using latency to evaluate interactive system performance. In Proceedings of the 1996.

No context found.

Y. Endo, Z. Wang, J. B. Chen, and M. Seltzer. Using Latency to Evaluate Interactive System Performance. In 2nd Symposium on Operating Systems Design and Implementation (OSDI, Seattle, WA, October 1996.

No context found.

Y. Endo, Z. Wang, J. B. Chen, and M. Seltzer. Using Latency to Evaluate Interactive System Performance. In 2nd Symposium on Operating Systems Design and Implementation (OSDI, Seattle, WA, October 1996.

No context found.

Yasuhiro Endo, Zheng Wang, Bradley Chen, and Margo Seltzer. Using latency to evaluate interactive system performance. In Proceedings of the 1996 Symposium on Operating System Design and Implementation, Seattle, WA, October 1996.

No context found.

Y. Endo et al., "Using Latency to Evaluate Interactive System Performance," Proc. 2nd Symp. Operating System Design and Implementation (OSDI '96), Usenix Assoc, Berkeley, Calif., 1996, pp. 185-199.

First 50 documents

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