Production File System SysCall Traces

This page provides a description of the traces that were used in the preparation of the research article titled “Non-blocking Writes to Files” that appeared in the Proceedings of the USENIX Conference on File and Storage Technologies, February, 2015.

1 day of traces collected during 09/16/2014. All the systems were running Linux using the ext4 and zfs file systems.

2 minute traces collected while replaying the original Mobibench traces. The replay was done on a Linux system using an ext4 file system.

The traces files (one per core) are in ASCII and each record is as follows:

In the case of READ or WRITE:

[seq num] [ts in ns] [operation] [inode num] [inode size in Bytes] [offset in Bytes] [size in Bytes] [HIT, MISS or APPEND]

• The inode size in Bytes is the size of the file before the write was issued.
• For ext4 accesses there is HIT, MISS or APPEND information. Such information is not available for ZFS accesses.
  • The HIT or MISS tag indicates if the accessed data was a HIT or MISS in the page cache of the system traced. When any part of the data accessed is not available in the page cache, we tag the request as a MISS.
  • The APPEND tag informs that the file size is increased beyond the page boundary of the last page. Small appends to the file that do not increase the file size to breach the page boundary are not tagged with APPEND.

In the case of OPEN, CLOSE, FSYNC, FDATASYNC operations, the format is different:

[seq num] [ts in ns] [operation] [inode num] [inode size in Bytes]