Summary

Lots of results! So what to take away?

• different I/O activity for different workloads!

• variation in block lifetime leads to variation in ideal write delay.

• small caches sharply reduce read traffic, larger caches don't necessarily increase portion of write traffic.

• memory-mapped is a much larger portion of file accesses and some of these files show much sharing amongst active processes.

• largest files are getting bigger, but they are not necessarily being accessed in a sequential manner reducing effectiveness of read-ahead.

• file access patterns are bimodal between mostly-read and mostly-write.

A major undertaking. A basis for motivating other work/improvements.