sql server data warehouse: I am trying to migrate timestamp fields in Teradata and
DB2 to SQL server. The problem is that Teradata and DB2
store timestamps to an accuracy of 1 microsecond, while
SQL Server stores it to the nearest 3.33 milliseconds
(the datetime datatype). The original accuracy needs to
be preserved as the columns are part of the primary key.

Is there an equivalent datatype in SQL Server or a quick
workaround for this problem? Date / Time functions need

There is no datatype in SQL Server that has a precision of 1 millisecond. I
can suggest 3 different workarounds:

- Store the timestamp in a CHAR(23) column in format
yyyy-mm-ddThh:mi:ss.nnn. You CONVERT this to and from datetime with style
126. Note that the precision will go back to 3.33 milliseconds when you
convert to datetime. This uses 23 bytes of storage.
- Store the timestamp as a UNIX timestamp (milliseconds since 1970-1-1) in a
BIGINT column. This uses 8 bytes of storage.
- Store the timestamp in 2 columns, one is a SMALLDATETIME (precision 1
minute), the second one is a INT, to store the number of milliseconds. This
uses 8 bytes of storage. (You can also use a SMALLINT if you want to save 2
bytes, but then you have to handle negative milliseconds.

Whichever is the most convenient of course depends on the calculations you
are going to do with it.

--
Jacco Schalkwijk
SQL Server MVP

[quoted text, click to view]