tomk@bert.ER.Bell.CA (Tommy Kanary) (11/21/90)
I have used Ingres's optimizedb program under Ingres 5.x and 6.x
on different Unix machines and I found
it to take more time to run then I expected.
For example : on a Sun 3/280 , 400K rows in a table, 3 sets of
secondary indexes, each set of indexes a composite of 3 cols
data types = varchar, maximum 21 characters per key. Only
optimizing on the index cols. Data table ~350 Mbytes,
secondary indexs 6 to 20 Mbytes each. Takes about 5 hours. ???
I also ran optimizedb on a 25 mip+ RISC machine on a 50 Mbyte
data base, 3 tables each with 20-50K rows and 70 cols. Storage type
is btree with a 5 col composite index of about 30 bytes. Ran against all
cols. Took 17 hours???
According to vmstat, on both these systems the cpu was the bottleneck.
Time was spent equally between system and user; context switches 1000+,
system calls 1000+. All cputime allocated against backend database and
optimizedb.
Does anyone know what operations optimizedb performs? and why should
these operations take so long? ( I can copydb in the databases and
and create indexs faster). Is this a program that Ingres has not "optimized"
or is this an inherently complex operation? Does it buy me enough in
performance to be worth the time to run the procedure?
--
Tom Kanary (416) 978-1303
Engineering & Research, Bell Canada
393 University Ave., Toronto, Ont, M5G-2E1
{utzoo}!censor!bert!tomk