Engineering topics

Iyer, Ravishankar K.

Publications and source records attributed to Iyer, Ravishankar K..

26 records · Page 2

NASA NTRS1990

Fault characterization of a multilayered perceptron network

The results of a set of simulation experiments conducted to quantify the effects of faults in a classification network implemented as a three-layered perception model are reported. The percentage of vectors misclassified by the classification network, the time taken for the network to stabilize, and the output values are measured. The results show that both transient and permanent faults have a significant impact on the performance of the network. Transient faults are also found to cause the network to be increasingly unstable as the duration of a transient is increased. The average percentage of the vectors misclassified is about 25 percent; after relearning, this is reduced to 10 percent. The impact of link faults is relatively insignificant in comparison with node faults (1 percent versus 19 percent misclassified after relearning). A study of the impact of hardware redundancy shows a linear increase in misclassifications with increasing hardware size.

Tan, Chang H.↗

NASA NTRS1989

Predictability of process resource usage - A measurement-based study on UNIX

A probabilistic scheme is developed to predict process resource usage in UNIX. Given the identity of the program being run, the scheme predicts CPU time, file I/O, and memory requirements of a process at the beginning of its life. The scheme uses a state-transition model of the program's resource usage in its past executions for prediction. The states of the model are the resource regions obtained from an off-line cluster analysis of processes run on the system. The proposed method is shown to work on data collected from a VAX 11/780 running 4.3 BSD UNIX. The results show that the predicted values correlate well with the actual. The correlation coefficient betweeen the predicted and actual values of CPU time is 0.84. Errors in prediction are mostly small. Some 82 percent of errors in CPU time prediction are less than 0.5 standard deviations of process CPU time.

Devarakonda, Murthy V.↗

NASA NTRS1988

Accurate low-cost methods for performance evaluation of cache memory systems

Methods of simulation based on statistical techniques are proposed to decrease the need for large trace measurements and for predicting true program behavior. Sampling techniques are applied while the address trace is collected from a workload. This drastically reduces the space and time needed to collect the trace. Simulation techniques are developed to use the sampled data not only to predict the mean miss rate of the cache, but also to provide an empirical estimate of its actual distribution. Finally, a concept of primed cache is introduced to simulate large caches by the sampling-based method.

Laha, Subhasis↗

NASA NTRS1987

Usage analysis of user files in UNIX

Presented is a user-oriented analysis of short term file usage in a 4.2 BSD UNIX environment. The key aspect of this analysis is a characterization of users and files, which is a departure from the traditional approach of analyzing file references. Two characterization measures are employed: accesses-per-byte (combining fraction of a file referenced and number of references) and file size. This new approach is shown to distinguish differences in files as well as users, which cam be used in efficient file system design, and in creating realistic test workloads for simulations. A multi-stage gamma distribution is shown to closely model the file usage measures. Even though overall file sharing is small, some files belonging to a bulletin board system are accessed by many users, simultaneously and otherwise. Over 50% of users referenced files owned by other users, and over 80% of all files were involved in such references. Based on the differences in files and users, suggestions to improve the system performance were also made.

Devarakonda, Murthy V.↗

NASA NTRS1987

Predictability of process resource usage: A measurement-based study of UNIX

A probabilistic scheme is developed to predict process resource usage in UNIX. Given the identity of the program being run, the scheme predicts CPU time, file I/O, and memory requirements of a process at the beginning of its life. The scheme uses a state-transition model of the program's resource usage in its past executions for prediction. The states of the model are the resource regions obtained from an off-line cluster analysis of processes run on the system. The proposed method is shown to work on data collected from a VAX 11/780 running 4.3 BSD UNIX. The results show that the predicted values correlate well with the actual. The correlation coefficient between the predicted and actual values of CPU time is 0.84. Errors in prediction are mostly small. Some 82% of errors in CPU time prediction are less than 0.5 standard deviations of process CPU time.

Devarakonda, Murthy V.↗

NASA NTRS1987

Recognition of error symptoms in large systems

A methodology for automatically detecting symptoms of frequently occurring errors in large computer systems is developed. The proposed symptom recognition methodology and its validation are based on probabilistic techniques. The technique is shown to work on real failure data from two CYBER systems at the University of Illinois. The methodology allows for the resolution between independent and dependent causes and, also quantifies a measure of the strength of relationship among errors. Comparison made with failure/repair information obtained from field maintenance engineers shows that in 85% of the cases, the error symptoms recognized by our approach correspond to real system problems. Further, the remaining 15% although not directly supported by field data, were confirmed as valid problems. Some of these were shown to be persistent problems which otherwise would have been considered as minor transients and hence ignored.

Iyer, Ravishankar K.↗

NASA NTRS1987

Measurement-based analysis of error latency

This paper demonstrates a practical methodology for the study of error latency under a real workload. The method is illustrated with sampled data on the physical memory activity, gathered by hardware instrumentation on a VAX 11/780 during the normal workload cycle of the installation. These data are used to simulate fault occurrence and to reconstruct the error discovery process in the system. The technique provides a means to study the system under different workloads and for multiple days. An approach to determine the percentage of undiscovered errors is also developed and a verification of the entire methodology is performed. This study finds that the mean error latency, in the memory containing the operating system, varies by a factor of 10 to 1 (in hours) between the low and high workloads. It is found that of all errors occurring within a day, 70 percent are detected in the same day, 82 percent within the following day, and 91 percent within the third day. The increase in failure rate due to latency is not so much a function of remaining errors but is dependent on whether or not there is a latent error.

Chillarege, Ram↗

NASA NTRS1986

Fault latency in the memory - An experimental study on VAX 11/780

Fault latency is the time between the physical occurrence of a fault and its corruption of data, causing an error. The measure of this time is difficult to obtain because the time of occurrence of a fault and the exact moment of generation of an error are not known. This paper describes an experiment to accurately study the fault latency in the memory subsystem. The experiment employs real memory data from a VAX 11/780 at the University of Illinois. Fault latency distributions are generated for s-a-0 and s-a-1 permanent fault models. Results show that the mean fault latency of a s-a-0 fault is nearly 5 times that of the s-a-1 fault. Large variations in fault latency are found for different regions in memory. An analysis of a variance model to quantify the relative influence of various workload measures on the evaluated latency is also given.

Chillarege, Ram↗