The queue data structure can keep data only in memory, only on disk
(writing it to disk and reading from disk), or in both (writing everything
to disk, but reading from disk only data stored in a previous run).
audisp-remote will use the last option for performance.

The queue file format starts with a fixed header, followed by an array
of slots for strings.  Due to the fixed size of each slot the file format
is rather inefficient, but it is also very simple.

The file is preallocated and the string slots will be aligned to a 4KB
boundary, so it should be necessary to only write one block to disk
when audisp-remote receives a (short) audit record.

With the default queue size of 200 items the file will be about 2.4
megabytes large, which is probably not really worth worrying about.

If necessary, the space utilization could be improved by storing strings
consecutively instead of using pre-arranged slots.

The queue file format is intended to be resilient against unexpected
termination of the process, and should be resilient against unexpected
system crash as long as the OS does not reorder writes (the string data
is written before the header that indicates that it is present) - but
ultimately resiliency against such failures is limited by other
links in the audit record transmission chain - if the record is lost
within auditd or audispd, having a resilient queue file format does
not help; audit records generated within the kernel are necessarily
lost if the system crashes before they are read by auditd because
the kernel will not be able to regenerate/retransmit them after the next
boot.