Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Statistics”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 235 records · Page 13
All sourcesNASA NTRSDOE OSTI
NASA NTRS1994

Long-term changes in the statistical distribution of Dobson total ozone in selected Northern Hemisphere geographical regions

The daily averages of total column amount of ozone taken in the period 1964-1988 at a network of 24 Dobson stations have been analyzed. Year-round data as well as summer data (May - Aug.) and winter data (Dec. - March) have been examined in the following regions: latitude bands (30 deg N - 39 deg N, 40 deg N - 52 deg N, 30 deg N - 60 deg N), North America, Europe, and Japan. To find year-to-year changes in the shape of the annual statistical distribution of total ozone (ASDTO) for these regions, we analyze trends in the following statistic characteristics of ASDTO: mean, standard deviation, median, and 10 and 90 percentiles. Time series of the statistical characteristics for the selected regions have been combined by averaging the individual stations values of these characteristics. The trends have been calculated by the multiple regression model adjusted for: the 11-year solar cycle, the Southern Oscillations effects, and for serial correlations. We have found that: a) in all regions (excluding Japan, North America), the shape of ASDTO has been drifting towards low ozone values. The drift seems to be not accompanied with a transformation in the shape of ASDTO. The drift speed (the rate of decrease in the annual means of total ozone) is of order 1-3 percent per decade (in the period 1970-1988). b) In Japan, the interannual changes in the shape of ASDTO have not been revealed. c) In North America, the drift of the year-round ASDTO (the year-round ASDTO comprises all the daily means of total ozone in a given year) has been accompanied with a transformation in the shape. The shape of the year round ASDTO becomes narrower. d) In all regions, except Japan and the band 30 deg - 39 deg N, the winter ASDTO (the winter ASDTO comprises the data taken in the period December in a given year through March next year) moves faster towards low ozone values than the summer ASDTO (the summer ASDTO comprises the data taken in the period May through August in a given year).

Krzyscin, Janusz W.↗
NASA NTRS1994

A method for obtaining a statistically stationary turbulent free shear flow

The long-term goal of the current research is the study of Large-Eddy Simulation (LES) as a tool for aeroacoustics. New algorithms and developments in computer hardware are making possible a new generation of tools for aeroacoustic predictions, which rely on the physics of the flow rather than empirical knowledge. LES, in conjunction with an acoustic analogy, holds the promise of predicting the statistics of noise radiated to the far-field of a turbulent flow. LES's predictive ability will be tested through extensive comparison of acoustic predictions based on a Direct Numerical Simulation (DNS) and LES of the same flow, as well as a priori testing of DNS results. The method presented here is aimed at allowing simulation of a turbulent flow field that is both simple and amenable to acoustic predictions. A free shear flow is homogeneous in both the streamwise and spanwise directions and which is statistically stationary will be simulated using equations based on the Navier-Stokes equations with a small number of added terms. Studying a free shear flow eliminates the need to consider flow-surface interactions as an acoustic source. The homogeneous directions and the flow's statistically stationary nature greatly simplify the application of an acoustic analogy.

Timson, Stephen F.↗
NASA NTRS1994

Redshift data and statistical inference

Frequency histograms and the 'power spectrum analysis' (PSA) method, the latter developed by Yu & Peebles (1969), have been widely employed as techniques for establishing the existence of periodicities. We provide a formal analysis of these two classes of methods, including controlled numerical experiments, to better understand their proper use and application. In particular, we note that typical published applications of frequency histograms commonly employ far greater numbers of class intervals or bins than is advisable by statistical theory sometimes giving rise to the appearance of spurious patterns. The PSA method generates a sequence of random numbers from observational data which, it is claimed, is exponentially distributed with unit mean and variance, essentially independent of the distribution of the original data. We show that the derived random processes is nonstationary and produces a small but systematic bias in the usual estimate of the mean and variance. Although the derived variable may be reasonably described by an exponential distribution, the tail of the distribution is far removed from that of an exponential, thereby rendering statistical inference and confidence testing based on the tail of the distribution completely unreliable. Finally, we examine a number of astronomical examples wherein these methods have been used giving rise to widespread acceptance of statistically unconfirmed conclusions.

Newman, William I.↗
NASA NTRS1994

Statistical association of QSO's with foreground galaxy clusters

We report a statistically significant overdensity of high redshift quasi-stellar objects (QSO's) in the directions of foreground galaxy clusters. QSO's are taken from the Large Bright QSO Survey (LBQS) between 1.4 less than or equal z less than or equal 2.2 with a limiting magnitude of m(sub B) = 18.5. Foreground clusters are regions within 6 Zwicky radii of small Zwicky clusters at a characteristic redshift of about z approximately = 0.2, covering about 40% of the total area surveyed (304 sq. deg). The overdensity, defined as the ratio of the number density of QSO's in the directions of clusters ('association QSO's) to that in the remainder of the fields ('background QSO's), is equal to 1.7, and formally differs from unity at 4.7 sigma significance. The observed overdensity probably is not due to statistical variation in QSO density, intrinsic QSO-QSO and/or cluster-cluster autocorrelations, or patchy Galactic obscuration. We thus interpret this observation as being due to statistical gravitational lensing of background QSO's by galaxy clusters. However, this amplitude of overdensity behind clusters cannot be accounted for in any cluster lensing model if the background QSO number-magnitude counts are similar to the intrinsic (unlensed) counts, and is implausible in any conventional model of cosmic mass distribution.

Rodrigues-Williams, Liliya L.↗
NASA NTRS1997

Development of Statistical Process Control Methodology for an Environmentally Compliant Surface Cleaning Process in a Bonding Laboratory

Bonding labs at both MSFC and the northern Utah production plant prepare bond test specimens which simulate or witness the production of NASA's Reusable Solid Rocket Motor (RSRM). The current process for preparing the bonding surfaces employs 1,1,1-trichloroethane vapor degreasing, which simulates the current RSRM process. Government regulations (e.g., the 1990 Amendments to the Clean Air Act) have mandated a production phase-out of a number of ozone depleting compounds (ODC) including 1,1,1-trichloroethane. In order to comply with these regulations, the RSRM Program is qualifying a spray-in-air (SIA) precision cleaning process using Brulin 1990, an aqueous blend of surfactants. Accordingly, surface preparation prior to bonding process simulation test specimens must reflect the new production cleaning process. The Bonding Lab Statistical Process Control (SPC) program monitors the progress of the lab and its capabilities, as well as certifies the bonding technicians, by periodically preparing D6AC steel tensile adhesion panels with EA-91 3NA epoxy adhesive using a standardized process. SPC methods are then used to ensure the process is statistically in control, thus producing reliable data for bonding studies, and identify any problems which might develop. Since the specimen cleaning process is being changed, new SPC limits must be established. This report summarizes side-by-side testing of D6AC steel tensile adhesion witness panels and tapered double cantilevered beams (TDCBs) using both the current baseline vapor degreasing process and a lab-scale spray-in-air process. A Proceco 26 inches Typhoon dishwasher cleaned both tensile adhesion witness panels and TDCBs in a process which simulates the new production process. The tests were performed six times during 1995, subsequent statistical analysis of the data established new upper control limits (UCL) and lower control limits (LCL). The data also demonstrated that the new process was equivalent to the vapor degreasing process.

Hutchens, Dale E.↗
NASA NTRS1998

Statistical Aspects of ENSO Events (1950-1997) and the El Nino-Atlantic Intense Hurricane Activity Relationship

On the basis of Trenberth's quantitative definition for marking the occurrence of an El Nino (or La Nina), one can precisely identify by month and year the starts and ends of some 15 El Nino and 10 La Nina events during the interval of 1950-1997, an interval corresponding to the most reliable for cataloging intense hurricane activity in the Atlantic basin (i.e., those of category 3-5 on the Saffir-Simpson hurricane scale). The main purpose of this investigation is primarily two-fold: First, the statistical aspects of these identified extremes and the intervening periods between them (called "interludes") are examined and, second, the statistics of the seasonal frequency of intense hurricanes in comparison to the extremes and interludes are determined. This study clearly demonstrates that of the last 48 hurricane seasons, 20 (42 percent) can be described as being "El Nino-related" (i.e., an El Nino was in progress during all, or part, of the yearly hurricane season--June-November), 13 (27 percent) as "La Nina-related" (i.e., a La Nina was in progress during all, or part, of the yearly hurricane season), and 15 (31 percent) as "interlude-related" (i.e., neither an El Nino nor a La Nina was in progress during any portion of the yearly hurricane season). Combining the latter two subgroups into a single grouping called "non-El Nino-related" seasons, one finds that they have had a mean frequency of intense hurricanes measuring 2.8 events per season, while the El Nino-related seasons have had a mean frequency of intense hurricanes measuring 1.3 events per season, where the observed difference in the means is inferred to be statistically important at the 99.8-percent level of confidence. Therefore, as previously shown more than a decade ago using a different data set, there undeniably exists an El Nino-Atlantic hurricane activity relationship, one which also extends to the class of intense hurricanes. During the interval of 1950-1997, fewer intense hurricanes occurred during El Nino-related seasons (always less than or equal to 3 and usually less than or equal to 2, this latter value having been true for 18 of the 20 El Nino-related seasons), while more usually occurred during non-El Nino-related seasons (typically greater than or equal to 2, having been true for 22 of the 28 non-El Nino-related seasons). Implications for the 1998 and 1999 hurricane seasons are discussed.

Wilson, Robert M.↗
NASA NTRS1999

Statistical Quality Control of Moisture Data in GEOS DAS

A new statistical quality control algorithm was recently implemented in the Goddard Earth Observing System Data Assimilation System (GEOS DAS). The final step in the algorithm consists of an adaptive buddy check that either accepts or rejects outlier observations based on a local statistical analysis of nearby data. A basic assumption in any such test is that the observed field is spatially coherent, in the sense that nearby data can be expected to confirm each other. However, the buddy check resulted in excessive rejection of moisture data, especially during the Northern Hemisphere summer. The analysis moisture variable in GEOS DAS is water vapor mixing ratio. Observational evidence shows that the distribution of mixing ratio errors is far from normal. Furthermore, spatial correlations among mixing ratio errors are highly anisotropic and difficult to identify. Both factors contribute to the poor performance of the statistical quality control algorithm. To alleviate the problem, we applied the buddy check to relative humidity data instead. This variable explicitly depends on temperature and therefore exhibits a much greater spatial coherence. As a result, reject rates of moisture data are much more reasonable and homogeneous in time and space.

Dee, D. P.↗
NASA NTRS2000

The GEOS Ozone Data Assimilation System: Specification of Error Statistics

A global three-dimensional ozone data assimilation system has been developed at the Data Assimilation Office of the NASA/Goddard Space Flight Center. The Total Ozone Mapping Spectrometer (TOMS) total ozone and the Solar Backscatter Ultraviolet (SBUV) or (SBUV/2) partial ozone profile observations are assimilated. The assimilation, into an off-line ozone transport model, is done using the global Physical-space Statistical Analysis Scheme (PSAS). This system became operational in December 1999. A detailed description of the statistical analysis scheme, and in particular, the forecast and observation error covariance models is given. A new global anisotropic horizontal forecast error correlation model accounts for a varying distribution of observations with latitude. Correlations are largest in the zonal direction in the tropics where data is sparse. Forecast error variance model is proportional to the ozone field. The forecast error covariance parameters were determined by maximum likelihood estimation. The error covariance models are validated using x squared statistics. The analyzed ozone fields in the winter 1992 are validated against independent observations from ozone sondes and HALOE. There is better than 10% agreement between mean Halogen Occultation Experiment (HALOE) and analysis fields between 70 and 0.2 hPa. The global root-mean-square (RMS) difference between TOMS observed and forecast values is less than 4%. The global RMS difference between SBUV observed and analyzed ozone between 50 and 3 hPa is less than 15%.

Stajner, Ivanka↗
NASA NTRS2001

Statistics of Low-Mass Companions to Stars: Implications for Their Origin

One of the more significant results from observational astronomy over the past few years has been the detection, primarily via radial velocity studies, of low-mass companions (LMCs) to solar-like stars. The commonly held interpretation of these is that the majority are "extrasolar planets" whereas the rest are brown dwarfs, the distinction made on the basis of apparent discontinuity in the distribution of M sin i for LMCs as revealed by a histogram. We report here results from statistical analysis of M sin i, as well as of the orbital elements data for available LMCs, to rest the assertion that the LMCs population is heterogeneous. The outcome is mixed. Solely on the basis of the distribution of M sin i a heterogeneous model is preferable. Overall, we find that a definitive statement asserting that LMCs population is heterogeneous is, at present, unjustified. In addition we compare statistics of LMCs with a comparable sample of stellar binaries. We find a remarkable statistical similarity between these two populations. This similarity coupled with marked populational dissimilarity between LMCs and acknowledged planets motivates us to suggest a common origin hypothesis for LMCs and stellar binaries as an alternative to the prevailing interpretation. We discuss merits of such a hypothesis and indicate a possible scenario for the formation of LMCs.

Stepinski, T. F.↗
NASA NTRS2000

Sampling Errors in Monthly Rainfall Totals for TRMM and SSM/I, Based on Statistics of Retrieved Rain Rates and Simple Models

Estimates from TRMM satellite data of monthly total rainfall over an area are subject to substantial sampling errors due to the limited number of visits to the area by the satellite during the month. Quantitative comparisons of TRMM averages with data collected by other satellites and by ground-based systems require some estimate of the size of this sampling error. A method of estimating this sampling error based on the actual statistics of the TRMM observations and on some modeling work has been developed. "Sampling error" in TRMM monthly averages is defined here relative to the monthly total a hypothetical satellite permanently stationed above the area would have reported. "Sampling error" therefore includes contributions from the random and systematic errors introduced by the satellite remote sensing system. As part of our long-term goal of providing error estimates for each grid point accessible to the TRMM instruments, sampling error estimates for TRMM based on rain retrievals from TRMM microwave (TMI) data are compared for different times of the year and different oceanic areas (to minimize changes in the statistics due to algorithmic differences over land and ocean). Changes in sampling error estimates due to changes in rain statistics due 1) to evolution of the official algorithms used to process the data, and 2) differences from other remote sensing systems such as the Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave/Imager (SSM/I), are analyzed.

Bell, Thomas L.↗
NASA NTRS2001

Linear and Order Statistics Combiners for Pattern Classification

Several researchers have experimentally shown that substantial improvements can be obtained in difficult pattern recognition problems by combining or integrating the outputs of multiple classifiers. This chapter provides an analytical framework to quantify the improvements in classification results due to combining. The results apply to both linear combiners and order statistics combiners. We first show that to a first order approximation, the error rate obtained over and above the Bayes error rate, is directly proportional to the variance of the actual decision boundaries around the Bayes optimum boundary. Combining classifiers in output space reduces this variance, and hence reduces the 'added' error. If N unbiased classifiers are combined by simple averaging. the added error rate can be reduced by a factor of N if the individual errors in approximating the decision boundaries are uncorrelated. Expressions are then derived for linear combiners which are biased or correlated, and the effect of output correlations on ensemble performance is quantified. For order statistics based non-linear combiners, we derive expressions that indicate how much the median, the maximum and in general the i-th order statistic can improve classifier performance. The analysis presented here facilitates the understanding of the relationships among error rates, classifier boundary distributions, and combining in output space. Experimental results on several public domain data sets are provided to illustrate the benefits of combining and to support the analytical results.

Tumer, Kagan↗
NASA NTRS2000

Statistical Study of Auroral Kilometric Radiation Fine Structure Striations Observed by Polar

We have conducted a statistical survey of a semirandom sample of the auroral kilometric radiation (AKR) data observed by the plasma wave instrument wideband receiver on board the Polar spacecraft. We have determined that AKR fine structure patterns with very narrowband, negative drifting striations occur in approximately 6% of the high-resolution wideband spectrograms when AKR is present. Positive sloping striations are also observed, but at a much lower rate. More than 8200 AKR stripes have been scaled. The stripes are predominantly found in the 40 to 215-kHz frequency range and have a frequency extent of about 4 kHz and a duration of usually less than 2 s. The majority of the stripes have drift rates between -8 and -2 kHz/s, with a peak in the distribution between -6 and -4 kHz/s. There is also a much smaller group of striations with positive drift rates of up to about 5 or 6 kHz/s. We have further investigated the change of drift rate with frequency. Almost all striations are observed in the lowest two frequency bands of the wideband receiver (f < 215 kHz). There is an increase in the statistical drift rate with increasing frequency. The statistical slope of the striations increases with frequency from about -4.4 kHz/s at 75 kHz to about -5.7 kHz/s at 170 kHz. This frequency dependence of the drift rate is consistent, under certain conditions, with a production mechanism stimulated by an upward propagating electromagnetic ion cyclotron wave, as had been suggested earlier. However, such a changing drift rate is also compatible with a stimulated source region that propagates upward along the magnetic field line at the velocity of an ion beam accelerated by a local, upward directed electric field, as is typically observed in the auroral region. An explanation for this association is not apparent at this time.

Menietti, J. D.↗
NASA NTRS1996

Statistical Models of Landscape Pattern and the Effects of Coarse Resolution of Satellite Imagery on Estimation of Area

Analysis of classified satellite imagery was conducted to characterize errors in estimates of area based on coarse resolution satellite imagery which are due to distortions in sizes of small fragments, and to explore the feasibility of correcting for these errors using a statistical modeling approach. Sizes of bodies of open water on ERS-1 SAR and fire scars on Landsat MSS imagery were measured. Statistical analysis of the smaller scars and ponds as observed with this imagery of relatively fine resolution demonstrated that the distribution of the sizes could be modeled by either of two types of statistical distributions - a power distribution related to fractal processes or a simple exponential distribution. Comparison of the distribution of small bum scars as observed with Landsat to the distribution observed with AVHRR showed distortions due to the coarse spatial resolution of AVHRR caused a net overestimation of bum area. This bias was primarily caused by detection in 2 or 3 AVHRR pixels of bums whose true size was on the order of an AVHRR pixel.

Hlavka, Christine A.↗
NASA NTRS1996

Constructing Space-Time Views from Fixed Size Statistical Data: Getting the Best of Both Worlds

Many performance monitoring tools are currently available to the super-computing community. The performance data gathered and analyzed by these tools fall under two categories: statistics and event traces. Statistical data is much more compact but lacks the probative power event traces offer. Event traces, on the other hand, can easily fill up the entire file system during execution such that the instrumented execution may have to be terminated half way through. In this paper, we propose an innovative methodology for performance data gathering and representation that offers a middle ground. The user can trade-off tracing overhead, trace data size vs. data quality incrementally. In other words, the user will be able to limit the amount of trace collected and, at the same time, carry out some of the analysis event traces offer using spacetime views for the entire execution. Two basic ideas are employed: the use of averages to replace recording data for each instance and "formulae" to represent sequences associated with communication and control flow. With the help of a few simple examples, we illustrate the use of these techniques in performance tuning and compare the quality of the traces we collected vs. event traces. We found that the trace files thus obtained are, in deed, small, bounded and predictable before program execution and that the quality of the space time views generated from these statistical data are excellent. Furthermore, experimental results showed that the formulae proposed were able to capture 100% of all the sequences associated with 11 of the 15 applications tested. The performance of the formulae can be incrementally improved by allocating more memory at run-time to learn longer sequences.

Schmidt, Melisa↗
NASA NTRS1997

Performance Data Gathering and Representation from Fixed-Size Statistical Data

The two commonly-used performance data types in the super-computing community, statistics and event traces, are discussed and compared. Statistical data are much more compact but lack the probative power event traces offer. Event traces, on the other hand, are unbounded and can easily fill up the entire file system during program execution. In this paper, we propose an innovative methodology for performance data gathering and representation that offers a middle ground. Two basic ideas are employed: the use of averages to replace recording data for each instance and 'formulae' to represent sequences associated with communication and control flow. The user can trade off tracing overhead, trace data size with data quality incrementally. In other words, the user will be able to limit the amount of trace data collected and, at the same time, carry out some of the analysis event traces offer using space-time views. With the help of a few simple examples, we illustrate the use of these techniques in performance tuning and compare the quality of the traces we collected with event traces. We found that the trace files thus obtained are, indeed, small, bounded and predictable before program execution, and that the quality of the space-time views generated from these statistical data are excellent. Furthermore, experimental results showed that the formulae proposed were able to capture all the sequences associated with 11 of the 15 applications tested. The performance of the formulae can be incrementally improved by allocating more memory at runtime to learn longer sequences.

Yan, Jerry C.↗
NASA NTRS2002

A Stochastic Model of Space-Time Variability of Tropical Rainfall: I. Statistics of Spatial Averages

Global maps of rainfall are of great importance in connection with modeling of the earth s climate. Comparison between the maps of rainfall predicted by computer-generated climate models with observation provides a sensitive test for these models. To make such a comparison, one typically needs the total precipitation amount over a large area, which could be hundreds of kilometers in size over extended periods of time of order days or months. This presents a difficult problem since rain varies greatly from place to place as well as in time. Remote sensing methods using ground radar or satellites detect rain over a large area by essentially taking a series of snapshots at infrequent intervals and indirectly deriving the average rain intensity within a collection of pixels , usually several kilometers in size. They measure area average of rain at a particular instant. Rain gauges, on the other hand, record rain accumulation continuously in time but only over a very small area tens of centimeters across, say, the size of a dinner plate. They measure only a time average at a single location. In making use of either method one needs to fill in the gaps in the observation - either the gaps in the area covered or the gaps in time of observation. This involves using statistical models to obtain information about the rain that is missed from what is actually detected. This paper investigates such a statistical model and validates it with rain data collected over the tropical Western Pacific from ship borne radars during TOGA COARE (Tropical Oceans Global Atmosphere Coupled Ocean-Atmosphere Response Experiment). The model incorporates a number of commonly observed features of rain. While rain varies rapidly with location and time, the variability diminishes when averaged over larger areas or longer periods of time. Moreover, rain is patchy in nature - at any instant on the average only a certain fraction of the observed pixels contain rain. The fraction of area covered by rain decreases, as the size of a pixel becomes smaller. This means that within what looks like a patch of rainy area in a coarse resolution view with larger pixel size, one finds clusters of rainy and dry patches when viewed on a finer scale. The model makes definite predictions about how these and other related statistics depend on the pixel size. These predictions were found to agree well with data. In a subsequent second part of the work we plan to test the model with rain gauge data collected during the TRMM (Tropical Rainfall Measuring Mission) ground validation campaign.

Kundu, Prasun K.↗
NASA NTRS2004

Statistical and Prediction modeling of the Ka Band Using Experimental Results from ACTS Propagation Terminals at 20.185 and 27.505 GHZ

With the increase in demand for wireless communication services, most of the operating frequency bands have become very congested. The increase of wireless costumers is only fractional contribution to this phenomenon. The demand for more services such as video streams and internet explorer which require a lot of band width has been a more significant contributor to the congestion in a communication system. One way to increase the amount of information or data per unit of time transmitted with in a wireless communication system is to use a higher radio frequency. However in spite the advantage available in the using higher frequency bands such as, the Ka-band, higher frequencies also implies short wavelengths. And shorter wavelengths are more susceptible to rain attenuation. Until the Advanced Communication Technology Satellite (ACTS) was launched, the Ka- band frequency was virtually unused - the majority of communication satellites operated in lower frequency bands called the C- and Ku- bands. Ka-band is desirable because its higher frequency allows wide bandwidth applications, smaller spacecraft and ground terminal components, and stronger signal strength. Since the Ka-band is a high frequency band, the millimeter wavelengths of the signals are easily degraded by rain. This problem known as rain fade or rain attenuation The Advanced Communication Technology Satellite (ACTS) propagation experiment has collected 5 years of Radio Frequency (RF) attenuation data from December 1993 to November 1997. The objective of my summer work is to help develop the statistics and prediction techniques that will help to better characterize the Ka Frequency band. The statistical analysis consists of seasonal and cumulative five-year attenuation statistics for the 20.2 and 27.5 GHz. The cumulative five-year results give the link outage that occurs for a given link margin. The experiment has seven ground station terminals that can be attributed to a unique rain zone climate. The locations are White Sands, NM, Tampa, Fly Clarksburg, MD, Norman, OK, Ft. Collins, COY Vancouver, BC, and Fairbanks, AK. The analysis will help us to develop and define specific parameters that will help system engineers develop the appropriated instrumentation and structure for a Ka-band wireless communication systems and networks.

Ogunwuyi, Oluwatosin O.↗
NASA NTRS1997

Constructing Space-Time Views from Fixed Size Statistical Data: Getting the Best of both Worlds

Many performance monitoring tools are currently available to the super-computing community. The performance data gathered and analyzed by these tools fall under two categories: statistics and event traces. Statistical data is much more compact but lacks the probative power event traces offer. Event traces, on the other hand, can easily fill up the entire file system during execution such that the instrumented execution may have to be terminated half way through. In this paper, we propose an innovative methodology for performance data gathering and representation that offers a middle ground. The user can trade-off tracing overhead, trace data size vs. data quality incrementally. In other words, the user will be able to limit the amount of trace collected and, at the same time, carry out some of the analysis event traces offer using space-time views for the entire execution. Two basic ideas arc employed: the use of averages to replace recording data for each instance and formulae to represent sequences associated with communication and control flow. With the help of a few simple examples, we illustrate the use of these techniques in performance tuning and compare the quality of the traces we collected vs. event traces. We found that the trace files thus obtained are, in deed, small, bounded and predictable before program execution and that the quality of the space time views generated from these statistical data are excellent. Furthermore, experimental results showed that the formulae proposed were able to capture 100% of all the sequences associated with 11 of the 15 applications tested. The performance of the formulae can be incrementally improved by allocating more memory at run-time to learn longer sequences.

Schmidt, Melisa↗