T Raceview: A Trace V Isua Ization Tool - UO Computer And Information .
T raceview:A TraceV isua izationToolALLEN D . MALONY,DAVIDDAVID J. ABLONOWSKl,H. HAMMERSLAG,andCenter for SupercomputingResearchand Development4b n ts thetrace-managementand I/O fea tunesusually found inspecial-purposeD-ace-analysissystems.Although they contain much performance detail, large tracefiles capturing logical or physical actionstaken by a program are difficult to usewhen analyzing a system’s perfomlance.The manual effort required to manipulatetrace files, including creating graphicalpresentations, can be daunting, requiringsome automatic support for trace analysisand visualization.Trace-based performance visualization gives you an intuitive understandingthat is often more useful than a textual statistical profile.‘.’ However, systems thatsupport performance visualization oftenonly accept trace input of a specific type,conformingto a particular executionparadigm or generated from a particularsystem context.‘-” Furthermore, the displays used can be inadequate to show thetime-dependentbehavior of arbitrarydata values that might be associated withdifferent events in the trace.’A general-purpose trace-visualizationsystem can incorporate common aspectsof trace processing and display. However,the gains in reusability come at the expense of specificity in trace-data analysis,because such a system must use a simplified event-interpretation model. Whetherthis trade-off is a liability will depend onthe trace-visualization application.In this article, we describe the design,development, and application of’T?aceview,a general-purpose n-ace-visualization tool.We seek to identify the aspects of trace visualization that can be incorporated into areusable tool and evaluate the trade-off ingeneral-purpose design versus semantically based, detailed trace-data analysis.ARCHITECTUREAND FUNCTIONThe architecture for a genera-purpose19
TracefibsDisplaystrace-visualization tool must be flexibleenough to let you select analysis and display alternatives. However, it should alsoprovide a structure rigid enough to let youbuild on the resources of the tool and extend the base analysis and display methods.Such an architecture cannot support alltrace-visualization models, but it shouldsupport most - especially those used forthe simple visualization problems thatoccur most frequently. Extension mechanisms should provide an easy customization path for more complex cases.We based Traceview’s general-purposearchitecture on the concept of a trace-visualization session. Figure 1 shows the hierarchical tree structure of a Traceviewsession, which involves trace files, views,and displays.First you specify a set of trace files tovisualize. For each trace file, you can define a set of views. A view defines a tracethe trace data, the number of definedviews, and the information necessary toreconstruct each view. The session manager assumes that a session-configurationfile is consistent with the data in the designated traces. In future implementations,the session manager will record the modification dates of trace files in the savedconfiguration and check the dates when itrestores the configuration.The session manager lets you mergemultiple session configurations into thecurrent session. If all trace files are distinct,merging is simple - basically, it is additive. When there are conflicting traceview combinations, you are prompted toresolve the conflict by selecting one alternative over another or by renaming entries.Figure 3 shows how a session appearsto a user. Open files are listed in the Fileswindow, defined views for the selected filein the views window, and created displaysfor the selected view in the Displays window. When vou select a trace file from theopen files list, Traceview automaticallyupdates the views list to show the corresponding defined views. Similarly, whenyou select a view, Traceview updates thedisplays list to show the view displays youhave created. You can add or delete files,views, and displays at any time. The number of trace files, view specifications, anddisplays is limited only by the memoryavailable to store the pertinent session information.subregion by setting a beginning and anending point, and by event filtering. Foreach view, you can create a set of displays.Although the session paradigm precludesdisplays that combine data from multipletraces, it supports multiple simultaneousdisplays. You can use multiple displays tocompare data from several trace files.Sessionmmopent. As Figure 2 shows,Traceview’s session manager savesand restores each session’s configuration andcoordinates the trace, view, and displaymanagers. The session manager lets yousave a session configuration to externalstorage for later retrieval of the same visualization environment. This saves workbecause ‘Ii-aceview visualization sessionscan be quite complex, with many tracefiles opened and many views defined oneach trace.At any point while you use Traceview,the session manager defines the currentsession as the set of open trace files, the setof defined views for each trace, and the setof displays for each view. However, thecurrent displays for each view are not partof the session configuration. They are defined only for the current Traceview invocation. Our initial reasoning was that saving the total display state would requiretoo much space. Now we are determiningif display saving can be accomplished withonly part of the display state.For each open trace, the session manager saves the name of the file containing167tracefile managerTracemanager ]*[DisployFigure 2. Tmcevim a &temwe.Luger 1UTrace fik. Traceview processes eventtraces. An event is a recorded instance ofsome logical action. We intentionally giveonly general descriptions of events because Traceview makes no semantic interpretation of the actions that the eventsrepresent. It assumes that each event istime-stamped merely to establish an ordering relation (usually a time ordering)among the events.The trace file is divided into two parts:an ASCII header and binary trace data,which is a time-sequenced list of traceevents. Each event reflects the instance ofsome action taking place during computation. Traceview interprets this action as astate transition. Each event recorded inthe trace includes encodings of the stateSEPTEMBER1991
- -“.I”“.““.“.“““,xI FILES“-l--l.DISPLAYS l”““ ,. .VIEWSDJ spl aySe1 ectediAl’s-.- -VI ewViewlype---.-l-”Canttscated when the trace was opened).The trace manager also provides agraphical user interface. You select tracesto open through a standard dialogue. Thetrace manager presents the list of open filesin the left window of the main ‘Ii-aceviewwindow, as shown in Figure 3, and presents summary information in the displayshown in Figure 4.- .- l.l.: jSe1 ectedDJ spl ay . . . . -- . “ .- “I”.” -.i dl spl ay-a,,OpeniNew1.I.-- . .“.I.“.“.“. .“. .Jl “ -.- --. .” ” .I. .- - .--being exited and the state being entered,an event type, and a time stamp. Eventsmay also include supplemental data fields.Events within a file are homogeneous intheir format: Each has the same number ofdata fields associated with it. (Future implementations will allow variable-size datafields dependent on the state.) Each datafield typically represents some numericevent metic.The trace-file header specifies thenumber of data fields associated with eachevent transition and how the data is labeled when presented to the user. It alsoprovidesflags to controldisplaycustomization, a directory of names to usefor the states, and an optional index intothe event data. Summary information inthe header includes the number of eventsin the trace and the total time representedby the trace data.Help -“- - ---Destroyi-.----- I -1- “I.“- ---View management.To define multipleviews on each trace, you use the view manager. A view deftition consists of a starting time in the trace, an ending time, and a list of names of events to be excluded from the trace.The view manager applies a view definition to a trace to produce a virtual aceid:1Start:0End .DataFields:8No.Views:3Trace mmagement.Working with thistrace-file format, the trace manager* opens trace files, interprets the trace-file header, calculates global trace statistics, reads events from open trace files,andl closestrace files (freeing storage allo-IEEESOFTWARE21
leName:Fi 1 ter- PLOTJRangeEvents-mmENDPLTXPANDMETRICAdjustLowerSe1 ectionAdJustBoundUpperBoundw (133(141(151INIT(161STEP(171Ti me :Time:Event:EventPatternFigure Y. Viewdefinition dialogue window. In the range selectionportion, you selector adjust a lower and upper bound oftime or- events. To jilter events,you toggleevent nanm individually or collectivelyin the eventfilter list. Andyou changethe eflect ofpattem matching using the Exclude and Include buttons.which is derived from the actual trace. Theview manager first discards any events thatoccur before the view starting time or afterthe view ending time. Then it filters theremaining events to remove the eventsyou specified for exclusion in the view definition.Figure 5 shows the user interface forview creation and modification. The viewrange is a lower and upper bound of timeor events, which you select or adjust. Forevent adjustments, the view managersearches for the named event forward orbackward from the current lower or upperbound. You make time adjustments eitherby entering a new time directly, or byusing the scroll bar. While it maintains itsnormal scrolling functions, which let usersmove a time window across the trace, thescroll bar can also be extended or contracted to change the time-window size.For event filtering, you toggle eventnames in the event filter list. You can toggle them individually or collectively, usingthe string pattern-matching capabilities.22IYou change the effect of the patternmatching using the Exclude and Includebuttons.Traceview performs event filteringbased on the view definition when constructing a virtual trace. Removing theevents that occur before or after the viewing range is trivial. Filtering out individualunwanted events is a bit more complicated. The view manager removes anevent from the trace if the event’s ‘I6 orFrom state designator matches an eventname specified for exclusion. The easiestway to understand the algorithm for eventfiltering is to see it as removing one eventat a time from the trace. At each iteration,the algorithm creates a new virtual tracewith all events to or from the state in question removed. The data fields are updatedappropriately. When it has removed allthe specified events, the final virtual traceis the result.Because trace files can be large, the costof reading raw traces entirely into memorycan be prohibitive. However, always read-ing trace events from secondary storagecan slow the system. Our approach inTraceview is to cache trace events in memory buffers. We chose to cache virtualtraces instead of raw traces because virmal-trace range selection and event filtering suppress events of no concern in thedisplay. However, Traceview does virmaltrace caching only for views, not for eachderivative view display.Traceview implements virtual-tracecaching by assigning each defined view acache buffer, the size of which you controlthrough a Traceview application resource.(In the future, we will add interactivebuffer-size control to the view window.)At any time, some consecutive portion ofthe virtual trace is cached. The time stampof the first cached event and the timestamp of the last cached event delimit theportion. If a display requests a virtual-traceevent within the cache, the view managerfetches the event from the cache, gettingthe benefits of fast memory access.However, if a display requests a virtual-traceI--SEPTEMBER1991
event outside the cache, the view managerflushes the cache and then refills it fromthe virtual-trace stream derived from theoriginal trace on secondary storage, starting with the requested event.Virtual-tracecaching minimizesTraceview’s total memory requirementswhile offering fast response time duringdisplay interaction. Speed depends on several parameters: the cache-buffer size, theview-range selection, the extent of eventfiltering, and the location of the virtualtrace references during display update.Further experience with Traceview, particularly with large trace files, will help usdetermine the efficacy of virtual-tracecaching.Display management.The display manager graphically presents the virtual traceconstructed by the view manager. The display manager lets you select horn twoavailable display methods and creates adisplay window showing the data. Thedisplay manager cormols the right window in Figure 3, which presents a list ofexisting displays for the selected view. Youcan reopen, destroy, and create displays.The display manager does not actuallydisplay the data; rather it invokes a displaymethod to display a virtualtrace.Traceview does not dictate a displaymethod. Although two display types arestatically linked with Traceview, you canincorporate any display method that caninterpret a Ti-aceview virtual trace.DISPLAYMETHODSThe display type menu lets you selecteither a Gantt display or a rate display.Both use the Gantt chart widget. Ganttcharts are line-plot representations oftime-sequencedperformancedata.Traceview uses a Gantt chart widget wedeveloped expressly for displaying macedata.Gantt chart widget. The Gantt chart widget provides horizontal and vertical axes,axes labels, data display, density bars, anddata averaging. In most cases,the numberof points displayed in a Gantt chart greatlyexceeds the pixel width of the x axis of theGantt chart. In our experience, the ratio ofdata points to pixels commonly exceeds10: 1. The Gantt chart widget offers twosolutions to this data-density problem:density bars and average curves.The display of a density bar on a chartis optional. A density bar is a band of colordisplayed above the Gantt’s data-displayarea. For density bars, Traceview lets youselect a color map, which assigns integerranges to colors. A density bar can represent either value density or point density.In a value-density bar, the color at pixel prepresents the average of all the data pointsdisplayed at pixelp on the x axis. In a pointdensity bar, the color at pixel p representsthe number of data points represented atpixel p on the x axis.In addition to a density bar and thegraphical data display, theGantt chart widget optionally displays an average curve. The averagecurve overlays the datadisplay and is computedby taking the average overa hxed interval of points asthe interval slides alongthe x axis. The averagedata value for point p onthe x axis is calculatedusing all the data pointsfrom the interval whencentered atp. If a point onthe x axis has no actualdata points associatedwith it, the Gantt chartwidget assumes it has the same averagevalue as the most recent point preceding it.axis. Because of this vertical stacking, thedensity and distribution of data points onthe x axis are identical for all the Ganttcharts in a given display shell.The display shell lets you control theindividual Gantt charts. From the displayshell you can add charts to or delete charts&om the display, mm density bars on oroff, turn averaging on or off, and producea Postscript version of the chart.The display shell also lets you manipulate all the Gantt charts together by adjusting the averaging interval and byzooming in on a chart region. You selectthe averaging interval by manipulating aslider to choose a percentage of the widthof the chart’s x axis. To zoom in on a chartportion, you use the mouse to select a region. Then you can zoom in on that region in all the charts. Youcan undo zooming step.wise or all at once.Althoughtwodisploy Using Gantt charts.typesarestatically Traceview uses the Ganttchart widget with the dislinkedwithTraceview,play shell to present twoof displays to theyoucanincorporate kindsuser: displays based onanydisplaymethod state transitions, whichwe call simply Gantt disthatcaninterpret plays,and displays of thenumber of times a state isa Traceviewentered, or ratesdi@zys.virtualtrace.For Gantt displays, theDisplayshell.Both display methods bundle Gantt charts in a display shell that synchronizes many Gantt charts in one window. The display shell also provides theuser interface where you control the behavior and appearance of the individualGantt charts.All the Gantt charts in a display shellderive from the same virtual trace, and thex axis for all the charts is identical. Therefore, the display shell stacks Gantt chartshorizontally and aligns them so a verticalline across the display shell intersects all itscharts at precisely the same point on the xchart’s x axis representstime. The y axis variesfrom chart to chart. For traces that contain no data fields, the display managershows only a single chart. Here, they axisrepresents states. A square wave showswhen a state is entered and when it is exited. If the trace also includes data fields,then the display shell includes a Ganttchart for each data field. For these, theyaxis represents the data field values. Because Traceview traces only have data recorded at state transitions, all the chartshave the same x axis and the same datapoint distribution.In the rates display, you view metricdata associated with a trace state. For theselected state, you can have a chart foreach data field deiined for the trace. TheIEEESOFTWARE23
tchesI/O MemoryReferencesCPU MemoryReferencesFloatingAddsPointing horizontal line displayed across theGantt chart.From within the display shell, you candump the details of a virtual mace to anASCII file. Traceview generates two typesof files. The first contains the details ofeach event record in the virtual trace, including time stamps and all the other fieldsof the event record. The second containsthe details of how Traceview constructedthe virtual ttace from the raw trace. IfTraceview excludes routines from the virtual trace, the most relevant data in this fileis how Traceview combined or ignoredevent records and their data fields to createthe virtual trace. We use the second type offile for diagnostics in virtual-trace generation.TRACEVIEWAPPLICATIONSFigure 6.disphy options.display manager also provides an additional chart giving the summation of allthe metrics available.For the rates display to be applicable,the trace must include data fields. In therates display, the x axis represents ordinalinstances of the state being entered. Theyaxis represents the individual data fieldvalues. Instead of the more familiar squarewave usually used in Gantt charts, the display manager causes the Gantt chart widget simply to display the points and connect them.MisceU0neOasfeatures. You can use themouse to place a horizontal line across theGantt chart plotting state transitions andfind out the name of the state representedby that point on they axis. Conversely, youcan bring up a list of all states in the virtualtrace, select one, and have the correspond-24A general-purpose trace-visualizationtool must be effective across a variety ofapplications. Using the tool’s standard features, you should be able to convenientlyprocess and visualize traces from real orsimulated systems. The visualizationshould help you gain insight into important performance phenomena difficult toobserve otherwise. The tool should let youinput trace data from different performance levels (hardware, system software,and application software) with differentdata appearing together in the trace. Finally, a trace-visualization environmentmust let you visualize and compare multiple traces simultaneously.lb see whether Traceview met thesepractical objectives, we used it to visualizetraces encountered in our performanceevaluation projects at the Center forSupercomputing Research and Development at the University of Illinois at Urbana-champaign. One project requiredTraceview to visualize program-eventtransitions and associated hardware performance information from applicationsprogramsrunningon Cray supercomputers. Another project was to determine via simulation the maximum parallelism within application executions. Inthis project, Traceview shows time-dependent levels of parallelism and corresponding system performance metrics.We elaborate on these two projects later.Another project not discussed here wasthe visualization of register- and functional-unit usage in traces from an instruction-level simulation of a single Cray XM P processor.We conclude from our validation teststhat Traceview is effective across differenttrace-visualization applications and requires little trace input modification, except for conversion to the standardTraceviewtrace format. AlthoughTraceview’s display methods are currentlylimited, the Gantt and rates displays offera robust visualization approach, paticularly for comparing different performancedata within a single trace or across multiple traces.We observed shortcomings in thosetrace-data 13nalysesthat required semanticinterpretation. In the future, we could addlimited semantic capabilities to Traceviewfor applications that require, for the mostpart, semantically independent trace analysis, but need some semantic trace interpretation. However, we will continue toemphasize trace-visualization applicationsin which the user provides the semanticcontext.ViewingGrayapplications.You can capturedetailed data about the performance behavior of an application’s execution onlyby tracing important events and samplingrelevant machine performance metrics.One tracing system for Cray supercomputers’ measures program-eventtransitions and stores information aboutmachine performance accessed from theCray hardware-performancemonit0r.aThere are four hardware performancemonitor groups, covering different classesof hardware performance, but only one ofthe four groups is accessible at a time.Tracing application execution can generate large amounts of trace data quickly,and analyzing and presenting the datamanually can be arduous. We have appliedTraceview to the analysis and display ofCray application traces. Here we describea sample Traceview session, where weviewed the trace from a vector execution ofthe Perfect benchmark code FL052 (amultigrid fluid-dynamics computation).’SEPTEMBER1991
The session display in Figure 3 is asnapshot of the session configuration usedfor the FL052 traces. The Files windowlists the four open trace files from differentvector executions of the FL052 program.The file ftrace contains an events-onlytrace, while the other traces include hardware-performancemonitor values foreach event in the trace. The ptrace filecontains the group 0 counter values. Thefiles trace02 and trace03 contain the group2 and group 3 traces, respectively. Figure 4shows the summary trace information forptrace.Because the ptrace trace contains bothevent data and hardware-performancemonitor data, you must be able to viewevent transitions and changes in hardwareperformance simultaneously. Then youcan analyze different computation regionsto see their effects on hardware performance. Figure 6 shows the Gntt Optionsmenu, which lets you choose interactivelythe data to be shown. Figure 7 shows theGantts display of the transitions betweenFL052 computational blocks during execution and the corresponding hardwarebehavior.Because the view definition for Figure7 selects all routines for viewing, all entryand exit transitions between the FL052routines appear in the Events chart. Allroutines are numbered and appear at theircorresponding number level on they axisin the Events chart. You can bring up theroutine-to-numbermapping in anotherwindow for review. Each vertical line inthe display represents a routine transition,and each horizontal line indicates the timespent in the current routine. Thus you canidentify the current routine at any point inthe display and observe the relative patternof routine occurrence and the differencesin routine execution time. For FLO52, theEvents chart shows the cyclic nature of themultigrid computation during one majorphase of the execution.Figure 7 also shows the values of thechosen hardware-performancemetricscorresponding to each routine transition.The displayed values are computed asrates over the period between successivestate transitions. In this case, the rates arein millions per second. From these dis-IEEESOFTWAREplays, you can see how hardware performance changes between routines. You canalso correlate performance across the different hardware metrics. The figure showsthat the cyclic nature of the routine transitions in the displayed region ofthe FL052computation is reflected in repetitivehardware behavior. This indicates relatively stable hardware performance withinand between the successive periods.A problem with displaying a largenumber of events is screen resolution. Thescreen reproduced in Figure 7 displays approximately 11,000 events in 900 pixels.The point-density bar in the Events display highlights areas where many eventsare presented. In this case, the color mapranges from black, through red, orange,and yellow, and finally to white. Black isthe least dense, white the most. The valuedensity bars in the various performancemetric displays show the average of performance values at the screen resolution.Zooming in on areas with high point density provides additional details about eventtransition and hardware performance.Figure 8 shows a zoomed-in Ganttsdisplay for the FL052 trace. Althoughthere is still an event-density problem, youcan discern some individual event transitions. This lets you observe and quantifythe high variability in discrete hardwareperformancetransitionsduring theFL052 computation. Clearly, the CPUMemory References metric is not uniformacross routines and shows significantchanges in the memory reference rate. Wealso observed this for the Total FloatingPoint Operations metric (not shown).You use the average curve, drawn overthe CPU Memory References metricchart, to contrast the discrete performance behavior with average performance. The CPU Memory Referencesmetric is a good example of howTraceview can simultaneously display performance data analyzed over differenttime intervals. Viewing detailed hardwaredata localizes the performance behavior toparticular routines, while averaged hardware data curves show aggregate performance over time.You use Traceview’s rates display toobserve the performance differences between successive invocations of a singleroutine. Figure 9 shows four performance
no resolution problem exists, you canquickly observe the cyclic nature of Psmooperformance in the view region. All Floating Point Operations and CPU MemoryReferences are positively correlated, butthey are negatively correlated with I/OMemory References. Floating Point Multiplies appear to be uncorrelated. The invocation sequence in a single period reflects the FL052 computation’s multigridnature. As the Psmoo routine is called onsuccessivelycoarser grids, the vector operations increase in overhead, resulting indecreases in delivered floating-point perFig-we 8. Zoomed FLOS2 Gantts disphy which letsyou discern mne individual event transitions. The CPUn/lemq Referencesmetric i.snot umfom acms routines a?zdsh . .ci l cal tchangesin the menmy ve ewzce formance and memory-reference rate.The main benefit of the rates displayvate. We aho obwwed t&for the Total Floating Point Operations metric (not shouz).for the FLOSZ application study is its ability to focus on a particular routine andmetrics for each invocation of the routine 1 FL052 code. A transition in the display display performance data for only thatPsmoo, a solution-approximationreflects a separate invocation in a time-orroutine. All the user-interaction capabilismoothingroutine in the multigrid d ere d seq uence of Psmoo calls. Because ties of the Gantts display are accessible /OMemoryReferences2.8546PSfUD3CPU MemoryReferences44.6809FloatingMultiplle -Figure 9. Fourpe omanre metricsfov eachinvocation of the routine Psnwo, a sobtion r pproximatio mloothing routine in the multigrid FLOJ2 rode.A transitionin the display rejects a separate invocation in a time-ordered sequenceof Prmoo calls. Th rates di.rpkzyfocuseson a particular routine and displaysperformance datafw only that routine.26SEPTEMBER1991
the rates display.We continue to use Traceview to studydifferent versions of Cray applications, including FLOSZ. Because Traceview lets usopen multiple trace files simultaneously,we can view traces from different executions on the screen at the same time. Thisis important in Cray work because wemust execute the programs several timesto capture traces for all four hardware-performance monitor groups. Also, we wantto compare the FL052 traces of vectorexecution with those from scalar and concurrent executions to better understandthe relative performance benefits of different compiler optimizations.Viewing max’hum parallelism.When evaluating the performance of a parallel computation, you want to know the maximumlevels of parallelism that could be achievedduring execution. Although in practice theinstantaneous parallelism will be limitedby the total processors available on themachine, a notion of peak parallelism canhelp you understand execution efficiency.Maxpar’” is a simulator for extracting aprogram’s maximum theoretically attainable parallelism. It works by maintaining aset of shadow variables for each actual variable. The shadow variable records thetime when the variable’s current value isvalid. Each operation in the original program, in addition to computing a changein the variable’s value, also computes thetime when that value is first avai
For each open trace, the session man- formation. ager saves the name of the file containing 1 Trace fik. Traceview processes event 67 trace file input session configuration saving and restoring Session manager Trace manager ]*[Disploy Luger 1 U Figure 2. Tmcevim a &temwe. traces. An event is a recorded instance of
STM is a trace element which allows for several types of trace messages to be added to a trace stream. This trace stream may or may not also contain instruction flow trace streams. Adding STM messaging has several effects. STM trace messages can be a way of tracing the high-level state of the SoC-based system.
1 2 3 5 4 Viv Lounge Chair 5 Trace 800 Coffee Table Circular Top Options Base Options Resources 3D visualisation models, 2D DWG and Revit files are available for all naughtone products Download the files from our website www.naughtone.com trace Small Table 1 Trace Coffee Table 2 Trace 800 Coffee Table Circular 3 Trace Side Table Images All images shown here and many others can be downloaded .
3.2. tips from our trace experience 28 3.3. evaluating your trace campaigns and use of tools 31 Using TAToo to analyse your data 35 Stories from the TRACE pilot sites 39 5.1. Traffic Snake Game 40 5.2. positive drive 44 5.3. biklio 50 TRACE recommendations 53 6.1. General recommendations 54 6.2.
APB200 - Advanced Debugging using the Ashling MPC5500 Tools Page 4 of 16 2. The e200 Code tab can be left at default settings i.e.: Figure 5. e200 Code Dialog 3. The e200 Data tab should be set as shown below. Set Start Emitting Data Trace to On Program Execution and Stop Emitting Data Trace to On Program Halt. Set Data Trace Region 0 Trace Type to Write Trace and the Start Address and End .
2.5 Installing Oracle Trace File Analyzer on Microsoft Windows in Non-Daemon Mode 2-4 2.6 Oracle Trace File Analyzer Key Directories 2-4 2.7 Oracle Trace File Analyzer Command Interfaces 2-5 2.8 Masking Sensitive Data 2-5 2.9 Securing Access to Oracle Trace File Analyzer 2-6 2.10 Uninstalling Oracle Trace File Analyzer 2-7iii
Nomenclature for Trace Element Classification . We have already grouped elements into two classes, major elements and trace elements (Figure 1). In addition, there are many ways of classifying trace elements. Examples are: 1) Siderophile elements: elements that concentrate in metallic iron such as Ni, Co, Os, Ir. Chalcophile elements
Directions: Answer these questions before you begin the Lab. 1. What are trace fossils? 2. How will you simulate trace fossils? Trace fossils can tell you a lot about the activities of organisms that left them. They can tell you how an organism fed or what kind of home it had. Real-World Question How can you model trace fossils that can
Proposed installation of underground storage tank (USTs) within groundwater protection zones (GPZs) has led to some conflict between the EA and developers in the past. Although standards for
