On Wed, 14 May 2003 04:51:01 -0700, Ron Hurley wrote:
> Attempt #3: Using parallel processes (17).
> Create VLTS.
> Move/Rebuild all objects from 34 of the largest
> tablespaces to VLTS. Most tables were moved within a
> couple hours, but after 2 days, the indexes were still
> being rebuilt. Server was extremely slow during the
> Index rebuild.
I used that method. For indexes, you should have used the PARALLEL clause
with nologging option. Also make sure that you have a big, realy big
SORT_AREA_SIZE (32MB or alike) and that your temporary tablespace reflects
that reality (extent of the temporary tablespaspace = SORT_AREA_SIZE+16k
for header. I don't know the exact size of a header by I've never had any
problems with 16k, which became my very own rule of thumb).
Also, monitor using "top", "glance" or "topas" that nobody is running CDE,
KDE, GNOME or alike and using 85% of the available CPU power, roughly half
of it in the kernel mode. I've done that with 0.8 TB 7.3.4 database on a
Siemens box and transferred it to 4-way HP cluster of 9000/N nodes. The
company is Oxford Health Plans and the project was called "PULSE
transition". It was completed on the Labor Day 2001. Also, if you have
OPS database, make sure that the parallel threads are not hitting the same
tablespace. Pinging will kill you. Hopefully, you did analyze what is
the index creation waiting for by using gv$session_wait and
gv$session_event?
--
Mladen Gogala
Software is like sex, it is better when it is free.
Linus Torvalds
Received on Wed May 14 2003 - 12:45:16 CDT
