Catálogo de publicaciones - libros

In this chapter we have tried to provide an overview of an important topic regarding the peculiar impact of self-similar Internet traffic on buffer sizing for routers and servers. The reason this is potentially very important for capacity planning is due to LRD clustering of Internet packets of the type first observed in the Bellcore measurements circa 1990. Such fractal-like clustering potentially leads to buffer overflow at much lower than conventionally expected traffic intensities.

Unfortunately, most of the details concerning LRD effects are contained in academic papers that are mathematically very sophisticated and impenetrable to the typical network capacity planner. This chapter has attempted to provide a simpler mathematical treatment than is generally available, but without any loss in accuracy. We concluded with some recent measurements and analysis that indicate the severity of these LRD effects may have been overestimated. Nonetheless, even if LRD effects are less important now than is currently portrayed, the wise GCaP planner will use the tools listed in Sect. 10.2 to monitor Internet traffic for their appearance in the future as multi-media payloads become more commonplace.

In this chapter we have tried to provide an overview of an important topic regarding the peculiar impact of self-similar Internet traffic on buffer sizing for routers and servers. The reason this is potentially very important for capacity planning is due to LRD clustering of Internet packets of the type first observed in the Bellcore measurements circa 1990. Such fractal-like clustering potentially leads to buffer overflow at much lower than conventionally expected traffic intensities.

Unfortunately, most of the details concerning LRD effects are contained in academic papers that are mathematically very sophisticated and impenetrable to the typical network capacity planner. This chapter has attempted to provide a simpler mathematical treatment than is generally available, but without any loss in accuracy. We concluded with some recent measurements and analysis that indicate the severity of these LRD effects may have been overestimated. Nonetheless, even if LRD effects are less important now than is currently portrayed, the wise GCaP planner will use the tools listed in Sect. 10.2 to monitor Internet traffic for their appearance in the future as multi-media payloads become more commonplace.

All performance measurements contains errors that need to be tracked to avoid wrong conclusions and misleading results. The Golden Rule in Sect. 3.4.1 states that the result of a calculation should not have more significant digits than the least precise number used. If you have absorbed the points of this chapter, you now know there are 3 sigdigs in the number 50.0 but only 1 sigdig in the number 50. You can verify that by manually applying Algorithm 3.1 or using the programs in Appendix D. The rules for rounding numbers to the appropriate sigdigs have been modified in recent times and the latest conventions are incorporated in Algorithm 3.2. The NIST guidelines at physics.nist.gov/cuu/Uncertainty/index.html present even more complex expressions of uncertainty in measurement.

Now, you also appreciate why CPU utilization data is never displayed with more than 2 sigdigs. If you see a CPU busy reading of CPU% = 20 output by your favorite performance monitoring tools, you now know that it really should be reported with an error margin along the lines of CPU% = 20 ± 5. Ignoring error margins leads to the sin of precision, or possibly worse!

In this chapter we have tried to provide an overview of an important topic regarding the peculiar impact of self-similar Internet traffic on buffer sizing for routers and servers. The reason this is potentially very important for capacity planning is due to LRD clustering of Internet packets of the type first observed in the Bellcore measurements circa 1990. Such fractal-like clustering potentially leads to buffer overflow at much lower than conventionally expected traffic intensities.

Unfortunately, most of the details concerning LRD effects are contained in academic papers that are mathematically very sophisticated and impenetrable to the typical network capacity planner. This chapter has attempted to provide a simpler mathematical treatment than is generally available, but without any loss in accuracy. We concluded with some recent measurements and analysis that indicate the severity of these LRD effects may have been overestimated. Nonetheless, even if LRD effects are less important now than is currently portrayed, the wise GCaP planner will use the tools listed in Sect. 10.2 to monitor Internet traffic for their appearance in the future as multi-media payloads become more commonplace.

In this chapter we have tried to provide an overview of an important topic regarding the peculiar impact of self-similar Internet traffic on buffer sizing for routers and servers. The reason this is potentially very important for capacity planning is due to LRD clustering of Internet packets of the type first observed in the Bellcore measurements circa 1990. Such fractal-like clustering potentially leads to buffer overflow at much lower than conventionally expected traffic intensities.

Unfortunately, most of the details concerning LRD effects are contained in academic papers that are mathematically very sophisticated and impenetrable to the typical network capacity planner. This chapter has attempted to provide a simpler mathematical treatment than is generally available, but without any loss in accuracy. We concluded with some recent measurements and analysis that indicate the severity of these LRD effects may have been overestimated. Nonetheless, even if LRD effects are less important now than is currently portrayed, the wise GCaP planner will use the tools listed in Sect. 10.2 to monitor Internet traffic for their appearance in the future as multi-media payloads become more commonplace.

In this chapter we have demonstrated that the universal scalability law, originally developed in the context of hardware capacity planning in Chaps. 4 and 5, is also applicable to software capacity planning. This is by no means obvious and a more formal queue-theoretic argument is presented in Appendix A. We applied the software version of the universal scalability law (6.7) to measurements based on the SPEC SDM benchmark on a UNIX platform, a database benchmark measured on a Windows NT platform, and the analysis of a multitier application.

For many readers, this version of the universal scalability law will most likely be the typical application. We termed this a virtual load-testing environment. In particular, some user loads of interest will lie beyond those achievable on the real test platform, either because the hardware configurations cannot be confiigured or the number of generators is limited by licensing costs.

We have examined the connection between concurrent programming, contention and coherency delays . Sect. 6.4 considered the thesis that the advent of multicore processors has brought concurrent programming back into the foreground for application development on multicores. We emphasized that this view can be made more quantitative in terms of (6.7) for concurrent-programming scalability.

Modern computing systems that abstract virtual resources from physical resources have surpassed the measurement paradigms of most performance management tools, thus they remain largely opaque to the performance analyst and capacity planner.

In this chapter, we introduced the spectrum of virtualization that spans microlevel VMs, e.g., hyperthreaded processors, through mesolevel hypervisors to macrolevel networked applications, e.g., GRIDs. Each of these regions has an identifiable proportional polling algorithm, and the polling frequency sets the location on the VM-spectrum. We used this framework to assess various performance case studies reported in the literature. It is clear from these studies that a lot of work remains to be done to better integrate VM performance instrumentation with current capacity management tools.

In this chapter we have tried to provide an overview of an important topic regarding the peculiar impact of self-similar Internet traffic on buffer sizing for routers and servers. The reason this is potentially very important for capacity planning is due to LRD clustering of Internet packets of the type first observed in the Bellcore measurements circa 1990. Such fractal-like clustering potentially leads to buffer overflow at much lower than conventionally expected traffic intensities.

Unfortunately, most of the details concerning LRD effects are contained in academic papers that are mathematically very sophisticated and impenetrable to the typical network capacity planner. This chapter has attempted to provide a simpler mathematical treatment than is generally available, but without any loss in accuracy. We concluded with some recent measurements and analysis that indicate the severity of these LRD effects may have been overestimated. Nonetheless, even if LRD effects are less important now than is currently portrayed, the wise GCaP planner will use the tools listed in Sect. 10.2 to monitor Internet traffic for their appearance in the future as multi-media payloads become more commonplace.

In this chapter we have demonstrated that the universal scalability law, originally developed in the context of hardware capacity planning in Chaps. 4 and 5, is also applicable to software capacity planning. This is by no means obvious and a more formal queue-theoretic argument is presented in Appendix A. We applied the software version of the universal scalability law (6.7) to measurements based on the SPEC SDM benchmark on a UNIX platform, a database benchmark measured on a Windows NT platform, and the analysis of a multitier application.

For many readers, this version of the universal scalability law will most likely be the typical application. We termed this a virtual load-testing environment. In particular, some user loads of interest will lie beyond those achievable on the real test platform, either because the hardware configurations cannot be confiigured or the number of generators is limited by licensing costs.

We have examined the connection between concurrent programming, contention and coherency delays . Sect. 6.4 considered the thesis that the advent of multicore processors has brought concurrent programming back into the foreground for application development on multicores. We emphasized that this view can be made more quantitative in terms of (6.7) for concurrent-programming scalability.

In this chapter we have tried to provide an overview of an important topic regarding the peculiar impact of self-similar Internet traffic on buffer sizing for routers and servers. The reason this is potentially very important for capacity planning is due to LRD clustering of Internet packets of the type first observed in the Bellcore measurements circa 1990. Such fractal-like clustering potentially leads to buffer overflow at much lower than conventionally expected traffic intensities.

Unfortunately, most of the details concerning LRD effects are contained in academic papers that are mathematically very sophisticated and impenetrable to the typical network capacity planner. This chapter has attempted to provide a simpler mathematical treatment than is generally available, but without any loss in accuracy. We concluded with some recent measurements and analysis that indicate the severity of these LRD effects may have been overestimated. Nonetheless, even if LRD effects are less important now than is currently portrayed, the wise GCaP planner will use the tools listed in Sect. 10.2 to monitor Internet traffic for their appearance in the future as multi-media payloads become more commonplace.