Use `strace` to analyze and optimize I/O sizes

Use strace to analyze and optimize I/O sizes (e.g., 8-byte reads when reading offsets?)

Summary

There may be suboptimal I/O patterns during operations such as archive reading or unpacking—particularly when reading offsets or metadata files. Preliminary observations suggest that the system may be issuing many small (e.g., 8-byte) reads, which can significantly degrade performance.

This issue tracks the use of tools like strace or perf to analyze actual system calls and I/O sizes in real-world workloads and guide improvements in buffer and read strategies.

Goals

• Identify inefficient I/O patterns (e.g., tiny reads, excessive seeks)
• Determine where in the code these are generated (e.g., reading offsets file or metadata)
• Optimize data access patterns (e.g., batch reads, buffer reuse)
• Improve throughput and reduce syscall overhead

Tasks

• Run strace -e trace=read,pread64 (or similar) on key operations like:
  • blk-archive unpack
  • blk-archive verify
• Look for repeated small reads (e.g., 8, 16, 32 bytes)
• Cross-reference read sizes with known structures (e.g., u64 offset entries)
• Propose buffered or chunked read strategies where appropriate
• Benchmark before and after I/O optimizations
• Ensure alignment with typical page/block sizes (4K, 64K)

Considerations

• Use larger, aligned buffers when possible to reduce syscall count
• Be mindful of memory usage when reading large slabs or metadata files
• Ensure that changes don't break forward/backward compatibility with slab/offset formats

Tools

• strace -e trace=read,pread64,openat
• perf record / perf report
• iotop, blktrace (for deeper block-layer insight)

Benefits

• Improved performance, especially on slow storage or under high load
• Reduced syscall and context-switch overhead
• Smoother behavior in high-throughput or I/O-constrained environments