If you have not yet read Stuart Carnie's article on slxdeveloper.com on using prepared queries with the SalesLogix provider than I highly recommend you check it out (See: The Myths and Legends of Prepared Queries). Stuart recently offered some comments to the article which provided some background on the provider in regards to trips to the server as well as other significant changes which result in big performance gains.
[...]6.1 always did a prepare before executing the statement. The main reason for removing this was scalability and performance. The load (both network and CPU) and memory usage on MSSQL is significantly greater than what is in 6.2, and preparing the query was a major contributing factor. Some early tests running my inhouse load testing tool showed a significant delta of memory usage for the sqlservr.exe process. 6.2 was about 65mb and 6.1.0 was about 240mb.
Round trips to the SQL server are also massively impacted. Almost always (after internal meta-data is cached), 6.2 performs 1 round trip to the server to execute the query, 6.1x performs 3, and depending on how the ADO recordset is used (i.e. if it is prepared and the executed), possibly even 5 round trips! Typically a 3 way is a 1:PREPARE, 2:EXECUTE, 3:UNPREPARE, a 5 is a 1:PREPARE, 2:UNPREPARE, 3:PREPARE, 4:EXECUTE, 5:UNPREPARE. This was largely to do with the SDK we used in 6.1x - 6.2 removed the SDK completely.
In a disconnected model, prepares rarely make any sense. Given our client is now disconnected, and ADO.NET is also a disconnected model, a prepare (by the SalesLogix provider) is even more unnecessary.
[...]Very rarely should code need to do a 'true' prepare. We have been working on this internally for 3+ years and have seen all the variations in performance. The balance we reached in 6.2 is the best for the SalesLogix application environment, including ADO.NET for how our developers including BPs typically use it.
[...T]he 6.2 provider truelly prepares the query in it's internal buffers, so it only parses, allocates parameter binding buffers and many other expensive internal structures once. The 6.1x provider (due to the 3rd party SDK we used) does all these steps EVERY time.
[...]Given the above query in the example is static, and only the parameter values change, SQL server caches the query anyway, and so the performance difference between doing a true PREPARE, EXECUTE..EXECUTE..EXECUTE, UNPREPARE is not worth the difference.
[See Stuart's full comments]