Hi,
Wonder if anyone has ever set-up striped raw disks for Oracle on AIX? I have
just down this for the first time, and have run into a problem...
- I've set-up my DB using striped raw partitions and everything starts and
runs fine. (You can't seem to switch off Mirror Write Consistency - but you
can turn
off Mirror Write Verify -for striped volumes. This may be part of the
problem?)
- I've created a tablespace for a particular table on one of the striped
raw partitions. I load 36,000,000 rows into it. The load goes fine. The ONLY
things on these disks are the stripes of the tablespace for this table,
NOTHING else.
- I try 'select count(*) from xt_fact' or select /*+ parallel(xt_fact,1) */
count(*) from xt_fact' and the process takes FOUR TIMES LONGER to complete
than normal - but ONLY the FIRST time you run it. Even if you shut the db
down and bring it back up, it's ALWAYS FAST on SUBSEQUENT occasions - in
other words, it's not an 'SGA caching the blocks issue'.
- If you watch the disks through iostat, it seems to WRITE as much as it
READS
for the above query when you run it for the FIRST TIME AFTER THE INITIAL
LOAD of data in the table! NB. There is NOTHING ESLE on the disks, just the
partitions for the above table in it's own tablespace. Why would it write?
- If you drop the table and reloaded it, then scan the table
with a parallel value of GREATER THAN 1, it's ALWAYS OK. The 'slow query'
and
'writing to disks problem' only ever shows itself when you query the table
NON-PARALLEL FOR THE FIRST TIME AFTER LOADING IT.
- The same problem exists even if you use filesystems (jfs) stripes instead
of raw stripes.
Anyone got any ideas?
TIA,
MotoX.
Received on Mon Jun 29 1998 - 03:02:34 CDT