The profiler will provide you with statistics on the execution time of requests and actions on the server. The total request time may or may not mean a lot, without linking them to specific tasks and query plans.
Other indicators of performance bottlenecks are resource competition counters (general statistics, latches, locks). You can see them through performance counters. In addition, searching for a large number of scan tables or other operations that do not use indexes may indicate the need for indexing.
On a busy server, an increase in parallelism is unlikely to significantly affect performance, since there are already many requests for a given time. Where parallelism gains on large rarely performed batch jobs, such as ETL processes. If you need to reduce the execution time of such a process, then parallelism can be a good place to watch. On a busy server performing a transactional load with many users, system resources will be occupied from the workload, so parallelism is unlikely to be a big win.
source share