Darshan

claude mcp add --scope project clio-darshan -- clio-kit mcp-server darshan

Install the launcher first. Restart your client after setup. Configuration locations →

clio-kit doctor --server darshan --connect

Tools​

load_darshan_log

Load and parse a Darshan log file to extract I/O performance metrics and metadata.

get_job_summary

Get job-level summary from a Darshan log including runtime, process count, and I/O volume.

analyze_file_access_patterns

Analyze file access patterns including read/write types and sequential vs random access.

get_io_performance_metrics

Extract I/O performance metrics including bandwidth, IOPS, and request sizes.

analyze_posix_operations

Analyze POSIX I/O operations including read/write system calls and their frequency.

analyze_mpiio_operations

Analyze MPI-IO operations including collective vs independent operations.

identify_io_bottlenecks

Identify I/O performance bottlenecks by analyzing access patterns and operations.

get_timeline_analysis

Report the job time span and each module's first/last open/read/write/close timestamps. Binned activity over time (peaks, idle periods) needs DXT trace data and is not computed from a Darshan summary log.

compare_darshan_logs

Compare two Darshan log files to identify performance differences between runs.

generate_io_summary_report

Generate a comprehensive I/O summary report with findings and recommendations.

Examples​

Replace example paths with your own files and check the returned results.

1. HPC Application Performance Analysis​

Analyze the I/O performance of my application using the Darshan log at /data/app_trace.darshan. Identify bottlenecks and provide optimization recommendations.

Tools called:

• load_darshan_log - Parse the Darshan trace file
• get_job_summary - Extract job-level statistics
• identify_io_bottlenecks - Find performance issues
• get_io_performance_metrics - Calculate detailed metrics

This prompt will:

• Use load_darshan_log to parse the trace file and extract metadata
• Generate job summary using get_job_summary for runtime and I/O statistics
• Identify performance bottlenecks using identify_io_bottlenecks
• Provide comprehensive performance analysis with optimization recommendations

2. I/O Pattern Optimization Study​

Compare the I/O patterns between /data/before_opt.darshan and /data/after_opt.darshan to validate our optimization efforts and measure performance improvements.

Tools called:

• analyze_file_access_patterns - Analyze access patterns for both files
• compare_darshan_logs - Compare performance metrics
• get_io_performance_metrics - Extract detailed performance data

This prompt will:

• Analyze access patterns using analyze_file_access_patterns for both traces
• Compare performance metrics using compare_darshan_logs
• Extract detailed metrics using get_io_performance_metrics
• Provide comprehensive optimization validation and improvement quantification

3. MPI-IO Collective Performance Analysis​

Examine the MPI-IO operations in /data/parallel_app.darshan, focusing on collective vs independent I/O patterns and their impact on overall performance.

Tools called:

• analyze_mpiio_operations - Analyze MPI-IO patterns
• get_timeline_analysis - Understand temporal patterns
• generate_io_summary_report - Create comprehensive report

This prompt will:

• Analyze MPI-I/O operations using analyze_mpiio_operations
• Generate temporal analysis using get_timeline_analysis
• Create detailed report using generate_io_summary_report
• Provide insights into collective I/O efficiency and optimization opportunities

4. POSIX System Call Analysis​

Investigate the POSIX I/O operations in /data/serial_app.darshan to understand file access patterns and identify potential optimizations for system call efficiency.

Tools called:

• analyze_posix_operations - Examine POSIX system calls
• analyze_file_access_patterns - Study file access behavior
• identify_io_bottlenecks - Find system-level bottlenecks

This prompt will:

• Analyze POSIX operations using analyze_posix_operations
• Study file access patterns using analyze_file_access_patterns
• Identify bottlenecks using identify_io_bottlenecks
• Provide system call optimization recommendations

5. Comprehensive I/O Performance Report​

Generate a complete I/O performance analysis report for /data/production_app.darshan including all metrics, visualizations, and recommendations for our production environment.

Tools called:

• load_darshan_log - Load and validate trace file
• generate_io_summary_report - Create comprehensive analysis
• get_timeline_analysis - Add temporal performance data

This prompt will:

• Load and validate trace using load_darshan_log
• Generate complete report using generate_io_summary_report
• Add timeline analysis using get_timeline_analysis
• Summarize I/O metrics and suggest optimizations to validate against the workload

Native backend configuration​

Native logs use darshan-parser --base. Request-size statistics are operation-weighted per-file averages, not individual requests. Job timing preserves subsecond precision. The timeline tool reports job duration; event-level peak and idle phases are not implemented.