Pompilio With regard to Properties vs UI, I’ll defer to others to answer as it is indeed debatable. With regard to best practices, there is no static answer, but i think you are on the right track when mentioning RAM,CPU etc. The way I would look at it in general is by looking ay what type of flow(s) you are running within a given instance, analyze them and see if higher levels of concurrency adds value since in a lot of cases adding more threads will actually kill the performance rather then give it a boost. For example, if your flow is CPU intensive then setting value higher then the amount of cores you have won’t produce anything good and in fact may have a negative impact on your performance as it may introduce unnecessary context switching. On the flip side if you are dealing with the flow(s) that is let’s say consist of a lot of IO intensive operations, then increasing concurrency most likely will help. So its a dance where the answer could also be as simple as saying “I want to limit NiFi to no more then 5 threads”.
That would be my 2cents Cheers Oleg > On Dec 2, 2015, at 11:20 AM, Pompilio Ramirez <[email protected]> wrote: > > All, > > As I was working on my flow I noticed that things worked slowly as I looked > around I noticed that the is a UI setting: > > "maximum timer driven thread count" > > I increased these numbers to 50 and my flows worked much better. > > I tried to look around but could not find more documentation on these setting > ( I could have overlooked them sorry ) > > I am wondering if there is a recommended setting? I cant remember default > value but increasing the setting on UI helped however not sure how to set > that at install .... > > I need to take in to consideration my RAM and CPU and my JVM max ram, but > just wondering what you all use as a benchmark? > > Also why is this setting only controlled by the UI, I assumed all settings > would be configured at the nifi.properties file and was surprised to find > this valuable nifi as a whole setting only configurable from the UI. > > > Thanks
