Let me at the start say that I find the new graphics very impressive. But I
am really concerned that Stata 8 is importantly slower than Stata 7. At
first I thought this was just in graphics, but using data is also slower.
Using Windows 2000, dual processors, SSCI drives, memory set to 200 MB,
here are some of the timings I am getting.
1) loading data into memory (use command).
Stata 7 313,132 obs = .74 second
Stata 8 313,132 obs = 7.47 second
Conclusion: Loading data into memory is 10 times slower
Same -use- statement plus some processing, sorting, preserving and data
clean up in a program (no graphics)
Stata 7 7.03 seconds
Stata 8 13.7 seconds
Conclusion: -use- is 10 times slower
Merging: Program with 5 merges and some data manipulation
Stata 7: 38.7 seconds
Stata 8: 33.4 Seconds
Conclusion merging and data manipulation is faster.
graph age pain 313,132 obs versus scatter age pain
Stata 7: 1.07 seconds
Stata 8: 46.6 seconds
44 times slower & a lot of seconds
spikeplot age, r(.5)
Stata 7: 2.02 seconds
Stata 8: 10.98 seconds
Five times slower - not too many seconds
xtgee xtgee haq age sex fatigue,i(pat) robust
Stata 7: 8.92 seconds
Stata 8: 9.20 seconds
About the same.
Overall conclusions and comments.
1) In general, using data (-use-) is a lot slower, but isn't really a much
of a problem.
2) New graphics are beautiful, comprehensive, and seem to be great.
3) However, when one wants a quick look at data rather than a perfect
graph, things really slow down and interfere with work, and this is
particularly so when you want multiple looks at data. Yes, I know use can
use graph7 (gr7), but it doesn't appear to be supported in that it isn't in
the manuals.
Suggestion. I recognize how much more complex the new graph system is and
that it just takes more time. But, would it be possible to add to the new
graph programs a new option (Quick) in which, if called, the program would
translate the new graph language into the old gr7 language and run a
"quick" graph. There would seem to be advantages to this approach. First,
it would make routine working easier and faster. Second, perhaps it would
enable one to get the graph language right without long delays. I often
find in Stata 7 that I have to run a graph 5-7 times to get it to look the
way I want it for publication. In the old 7 system that would be 8 seconds
of processing, but in the new Stata 8 system it would be > 5 minutes.
Third, it wouldn't require the support of the old language (just its sotto
voce implementation).
One wonder if Stata Corp might be able to comment on these speed issues and
suggestion.
Thanks,
Fred
----------------------------------------------------------------------------
------------------------
Fred Wolfe Tel
(316) 263-2125
National Data Bank for Rheumatic Diseases Fax (316) 263-0761
Wichita,
Kansas [email protected]
----------------------------------------------------------------------------
-------------------------
