Running DB2/UDB 8.1 on Linux, 16G RAM, 8 processors. We are currently
converting data from legacy systems and after importing large amounts
of data into multiple tables, I run a script that executes runstats on
all columns with distribution on all columns and detailed indexes for
every table in the database. At first, I had the default setting for
stat_heap_sz, but increased it to 20000 and then 100000 and still
haven't seen any improvement in speed. Currently, it takes about 2
hours to execute runstats. Any suggestions for increasing the speed?

Thanks,

Chris

cbcobb@gmail.com wrote:[color=blue]
> Running DB2/UDB 8.1 on Linux, 16G RAM, 8 processors. We are currently
> converting data from legacy systems and after importing large amounts
> of data into multiple tables, I run a script that executes runstats on
> all columns with distribution on all columns and detailed indexes for
> every table in the database. At first, I had the default setting for
> stat_heap_sz, but increased it to 20000 and then 100000 and still
> haven't seen any improvement in speed. Currently, it takes about 2
> hours to execute runstats. Any suggestions for increasing the speed?[/color]

STAT_HEAP_SZ doesn't affect performance.

Runstats will basically do table and index scans, so you want to
tune your bufferpools and prefetching. Look at your tablespace layout
and how the tablespace containers are placed on your disk drives.

Also you could consider using sampling via the TABLESAMPLE and SAMPLED
INDEXES option.

Or, (make IBM happy) and license the database partitioning feature. :-)

<cbcobb@gmail.com> wrote in message
news:1124725588.787604.289380@g47g2000cwa.googlegr oups.com...[color=blue]
> Running DB2/UDB 8.1 on Linux, 16G RAM, 8 processors. We are currently
> converting data from legacy systems and after importing large amounts
> of data into multiple tables, I run a script that executes runstats on
> all columns with distribution on all columns and detailed indexes for
> every table in the database. At first, I had the default setting for
> stat_heap_sz, but increased it to 20000 and then 100000 and still
> haven't seen any improvement in speed. Currently, it takes about 2
> hours to execute runstats. Any suggestions for increasing the speed?
>
> Thanks,
>
> Chris
>[/color]
Most likely you do not need distribution on all columns. Key columns is
enough. This is especially true now that you have run it once with all
columns.