From: 101635.2725@compuserve.com (Riaz Alam)

Ignatius Fernandez 
wrote:

>Document US-11869 in Ingres Advisor reads as follows:

>>  Currently in 6.3 rollforwarddb is a stand alone utility which
>>  uses the base table access method to access the table.  If
>>  the base table is a 2-3 million row heap with btree secondary
>>  indices, rollforwarddb will still use the heap access method
>>  to access that table.                                                         
>>
>>  In this case appends are still fast, but replaces or updates
>>  (inserts) may take ~15 minutes each.  Resolution is to put an
>>  index structure on the base table. Future release will institute
>>  logical logging in which alternate access methods would be utilized.

>Does the above continue to hold true for 6.4 and OpenIngres 1.x?
>I recollect a presentation at Ingres World 96 where the issue came up.

>Thanks.
>Ignatius.

The same applies to 6.4 but OpenINGRES 1.x doesn't perform full table
scans to apply records when rolling forward heap tables.  It's a damn
site better than 6.4.

The recovery algorithms have been redesigned in OI 1.x and are common
between the recovery server and rollforwarddb.  I tested a rollforward
of an update which affected 4000 rows and it was 400% faster!  The
table was a btree with isam secondaries.

Riaz

Ingres Q & A

To William's Home Page

© William Yuan 2000

Email William