Ray:

Regarding smaller "secondary" histories for the timesgcm,
suppose I add the following lexical reads:

SECOUT  = list of mss file paths for secondary histories 
          (analogous to OUTPUT)
SECHIST = model time interval at which to save secondary histories
          to disk (analogous to HIST) (if SECHIST = 0,0,0 then 
          secondary histories would not be written at all, otherwise
          it must be a multiple of STEP)
SECSAVE = model time interval at which to dispose secondary history
          file(s) to mss (analogous to SAVE) (if SECSAVE=0,0,0 then
          secondary histories would not be disposed to mss, otherwise
          it must be multiple of SECHIST)

The main question here is how many and which fields are to be written
to the secondary histories. Here are 3 possibilities:

1) Write all fields for the current time on the secondary histories, 
   leaving off the fields from the previous time step (both current 
   and previous fields are written on the big restart histories). This 
   would reduce the size of a history by about 1/3 (29 instead of 43 
   fields), allowing say 6 histories per volume rather than the current 
   4. So the secondary histories would still be pretty big, but you 
   would have all the fields, and no changes would need to be made to 
   any of the processors(!).

2) Hardwire only the fields you want, say t,u,v,w,o2,o,n2. The
   histories would be quite small, and I would have to alter the 
   processors accordingly (they could read a flag in the header
   to know whether its a primary or secondary history)

3) Add a 4th new lexical read in which you would specify which
   fields are to be written to the secondary histories. I would record
   that information in the header so the processors would know how much
   to read and what could be plotted. This would be the most generic
   option, but probably also the most complicated (secondary histories
   would vary in size).

--Ben