Due to the design of indexes within a dataserver that provides an optimum search path, index rows are ordered for fast access to the table's data row. Index rows which contain row identifiers (RIDs), are treated as binary to achieve a fast access to the user table.
Within the same architecture platform, the order of index rows remains valid and search order for a selection criteria takes its normal path. However, when index rows are translated across different architectures, this invalidates the order by which optimization was done. This results in an invalid index on user tables when the cross platform dump and load feature is performed.
A database dump from a different architecture, such as big endian to little endian, is loaded, certain indexes are marked as suspect:
Non-clustered index on APL table.
Clustered index on DOL table.
Non-clustered index on DOL table.
To fix indexes on the target system, after load from a different architecture dump, you could use one of two methods:
Drop and recreate all of the indexes.
Use sp_post_xpload, see page §.
Since the data point and information varies from usage on indexes, the schema, user data, number of indexes, index key length, and number of index rows, in general, it requires planning to recreate indexes on large tables as it can be a lengthy process. sp_post_xpload validates indexes, drops invalid indexes, and recreates dropped indexes, in a single command on databases.
Since sp_post_xpload performs many operations it can take longer than drop and recreate indexes. Sybase recommends that you use the drop and recreate indexes on those databases larger that 10G..
| Copyright © 2005. Sybase Inc. All rights reserved. |
|
|
