I’m trying to use the Task-Parallel-Library to offload expensive ADO.NET database access from the UI thread (formerly the program I’m re-writing would simply freeze, occasionally updating a VB6 text box with its progress, until the data in the database was fully loaded). I have an complex dependency structure (26 individual tasks), and I’m trying to figure out how much of it is worth parallelizing.
I’d like to know whether or not IO access like this can be parallelized at all with performance bonuses. If not I’ll just sequentially load the data and update the UI whenever enough information is loaded to perform that task, but It’d be nice to get an extra boost by loading maybe two things at a time instead of just one (even if I don’t get double speedup).
It’s possible that parallelizing this will increase performance, but not guaranteed. It all depends on where your bottleneck is.
For example, if a request is expensive because it loads lots of data, then it probably consumes much of your clients network bandwith. Parallelizing in this case wouldn’t help much, if at all.
If, on the other hand, the bottleneck is the SQL processing and your SQL request leaves the SQL Server with spare capacity in its own bottleneck, then you can profit from SQL Servers (very good) parallelizing capabilities.
It is also possible that parallelizing slows you down. If for example the SQl Server has not much RAM and access only to a single disk, forcing it to do multiple queries in parallel may lead to more seek activity on the harddisk, which can dramatically slow down the overall read rate.
So, as it often is, the answer isn’t a simple yes or no, but “it depends”.