Archives of the TeradataForum

Message Posted: Mon, 25 Oct 2004 @ 15:50:37 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

Fallback clustering is set up to minimize the shared failure modes of the amps in each fallback cluster. So for example, in a multi-clique system, no two fallback cluster mates are in the same clique. This then eliminates any shared failure mode in the storage hardware, since cliques don't share storage resources with each other. Even in a single clique system, cluster mates are placed on different disk arrays if possible.

The three performance measurements relevant to the Teradata Disk I/O Integrity Check are the following:

1. The primary performance measurement for checksum characterization is increased CPU path length per transaction, which is caused by the time required to calculate and verify a checksum.

2. A secondary measure of performance is increased I/O count per transaction, which is caused by no longer updating in place for data on which checksums are enabled.

3. A third measurement for performance is throughput which is directly related to the elapsed time required to complete a transaction.

A set of full table modification tests were run in order to measure the performance. The full table modification tests operated on a large number of rows, and include insert, delete, and update operations on 1%, 10%, and 30% of the rows in the table. Additionally, utilities like FastLoad, MultiLoad, and Tpump were executed.

Keep in mind that the following results would only apply to the data being checksummed. Also, of course, your mileage may vary.

CPU Path Increase

Checksumming 2% of the data in each disk sector

          Deletes 0%
          Inserts 0% - 5%
          Updates 13% - 25%

Checksumming 100% of the data in each sector

          Deletes 13% -41%
          Inserts 20% - 64%
          Updates 34% - 101%

I/O Count Increase

Checksumming 2% of the data in each disk sector

          Deletes 0%
          Inserts 0%
          Updates 0% - 6%

Checksumming 100% of the data in each sector

          Deletes 0% - 2%
          Inserts 0% - 2%
          Updates 3% - 6%

Throughput decrease

Checksumming 2% of the data in each disk sector

          Deletes 0% - 2%
          Inserts 0% - 1%
          Updates 1% - 8%

Checksumming 100% of the data in each sector

          Deletes 5% - 6%
          Inserts 4% - 9%
          Updates 6% - 12%

Mark

	<Prev	Next>		<<First	<Prev	Next>	Last>>