Archives of the TeradataForum

Message Posted: Fri, 22 Oct 2004 @ 19:18:07 GMT

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

Actually the hardware checksum (or the cyclic redundancy check - CRC) is always verified when any sector is read. If there is a CRC error (or parity error - not necessarily the same thing), it's reported to the operating system at the time of the read as an unrecoverable error. It's up to the operating system to either act upon or escalate the error.

In the case of your example, where the data has been written correctly to disk and then later corrupted due to a hardware fault, the hardware CRC should detect the error at the time of read. It doesn't matter if one sector (or twenty) are read, if there's been corruption at the physical level, then the fault should be detected by the hardware and the bad sector(s) reported when read. If you can't trust what is happening at the physical level, then how can you trust anything dependent on the hardware?

What about the problem where the RDBMS or the Operating System has corrupted the data before it's written? The corrupted data will be written with a valid CRC and the corruption will go undetected (either during the write or by the later read).

Where SCANDISK checks for the physical error, CHECKTABLE looks for logical problems (ie- missing fallback rows or table structure errors). If the data was corrupted before it was written, it will probably pass SCANDISK and depending on the nature of the corruption, it might fail CHECKTABLE.

So if the CHECKSUM option is being handled at the RDBMS level and the corruption is caused before the data is written, how is the CHECKSUM option going to be any more effective than the CRC written by the disk subsystem? And being the hardcore cynic, I can't help but wonder that if the CHECKSUM option is a good idea for tables, wouldn't it also be a good idea for spool files? If I can trust the spool files, then why can't I trust the tables?

So that's why I say I find the need for the CHECKSUM option a little confusing.

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