Back-of-the-envelope cost of extra data :-)

Robbin Kawabata Robbin.Kawabata at eng.sun.com
Mon May 9 21:09:53 UTC 2005


Amending my previous email of 5/3/05 (where I said either method is OK),
our preference is the algorithmic approach.

Here are some comparisons (using the smaller 5-byte time storage proposed
5/9/05 by ado).

The 400-year method requires in a larger timezone file (was 1K, now 6K).
localtime() requires 3K/5K (32/64 bit) more memory per process for
the state struct.  There is more setup processing to initialize 400-years
of the ats[] table.  It requires about .8MB more hard disk space for the total
zoneinfo directory.  All processes encounter the overhead, but most processes
would not make use of it.  The performance for 64-bit transitions is
probably better, as it is table-driven.

The algorithmic approach requires a smaller timezone file than the 400-year
approach.  Some number of extra bytes are needed to store the variables to
be used by the algorithm (and more space needed for a small 64-bit transition
table, if one is used).  There is a little setup for initialization of
variables that would be used by the algorithm (and more setup for the small
64-bit transition table, if one is used).  Additional hard disk space is
minimal (or somewhat less than the 400-year file if a small 64-bit transition
table is used).  Performance for 64-bit transitions (past the 64-bit transition
table if one is used) would probably not be as good as the 400-year cycle
approach.

We prefer the algorithmic method because it requires less overhead with disk
space, process memory space, and setup processing time.  We feel the possible
slower performance for 64-bit transitions is acceptable since most
applications won't use 64-bit transitions.

Robbin




More information about the tz mailing list