Seeing Through The Bottleneck

13
+ Seeing Through The Bottleneck The Vizen Trace Visualization Tool Matthew Pruitt, Jeremiah Barr Final Report: Graduate Operating Systems Vizen Conceptualize Visualize Recognize Optimize

description

Conceptualize. Visualize. Vizen. Recognize. Optimize. Seeing Through The Bottleneck. The Vizen Trace Visualization Tool Matthew Pruitt, Jeremiah Barr Final Report: Graduate Operating Systems. The Problem. The tool strace Copious amounts of data - PowerPoint PPT Presentation

Transcript of Seeing Through The Bottleneck

Page 1: Seeing Through The Bottleneck

+

Seeing Through The BottleneckThe Vizen Trace Visualization ToolMatthew Pruitt, Jeremiah BarrFinal Report: Graduate Operating Systems

Vizen

Conceptualize Visualize

Recognize Optimize

Page 2: Seeing Through The Bottleneck

+The Problem The tool

strace Copious amounts of data Difficult to read for users unfamiliar with the strace

syntax top

Hard to organize processes and limit the view Doesn’t relate different performance measures

Visualizing The Trace High dimensionality in the trace Aggregating data from trace and other processes can cause

high overheadopen("/lib64/libdl.so.2", O_RDONLY) = 3read(3, "\177ELF\2\1\1\0\0\0\0\0\0\0\0\0\3\0>\0\1\0\0\0\20\16\30040\0\0\0"..., 832) = 832fstat(3, {st_mode=S_IFREG|0755, st_size=23360, ...}) = 0mmap(0x3034c00000, 2109696, PROT_READ|…

Page 3: Seeing Through The Bottleneck

+The Solution

Abstraction Create multiple views showing key points in the data

Allows users to see connections between trace data and various aspects of system performance

Users can drill down to pertinent information Allows user to easily to aggregate data that would

normally be provided in strace

Difficulties Measuring statistics nearly concurrently Maintaining low overhead Displaying data in such a way that the user can quickly and

intuitively see bottlenecks

Page 4: Seeing Through The Bottleneck

+/proc Virtual File System

/proc/PID fd

stat

Folder containing symbolic links to open files

File containing various statistics about a program, including page faults and CPU utilization

Page 5: Seeing Through The Bottleneck

+The Diagram of The Solution

JBarChartJXYPlot JGlobalView

JTraceDataModel

JProcExplorer

JPerfDataModel

JProcessParser

Vizen

strace/proc VFS

Page 6: Seeing Through The Bottleneck

+Evaluation

Page 7: Seeing Through The Bottleneck

+Evaluation (cont’d)

Page 8: Seeing Through The Bottleneck

+Evaluation (cont’d)

Page 9: Seeing Through The Bottleneck

+Evaluation (cont’d)

Page 10: Seeing Through The Bottleneck

+Evaluation (cont’d)

Page 11: Seeing Through The Bottleneck

+Effects on RuntimeMethod Mean Execution TimeBaseline 43.4Top 44.0Strace 160Vizen 56.7

Page 12: Seeing Through The Bottleneck

+Future Work

Still need to combine all views Implement text filtering Performance tuning

Page 13: Seeing Through The Bottleneck

+The End