Again, it depends on your topology how long it would take for the latency numbers to stabilize. I cannot answer that for you. In general, the "complete latency" is a pretty good proxy for the behavior of the topology. If it's lower that's probably better.
The capacity might be *related* to the CPU utilization (but not necessarily, again, there are too many factors). If you want to measure CPU utilization, you should actually measure the CPU utilization on the hosts directly instead of trying to infer it from metrics like capacity. Answering the Q of "how do I measure CPU utilization" is probably outside of the scope of this list. That's a very general sysadmin / dev-ops type of question, to which there is no single solution. - Erik On Mon, Mar 21, 2016 at 5:17 PM, sam mohel <[email protected]> wrote: > Thanks to you I got it . but what should I do when I need to compare > between two topologies > > I read about tools like ganglia that gave graph about CPU utilization > should I use it to make measuring performance easier than numbers > > And is there specific columns in storm ui that should be greater than > other or less than other ? > > Last question i read that I can measure CPU utilization by capacity column > in storm ui > But don't know what are numbers should I get . is there specific number > that I can compare with it ? > > Sorry for my lot questions . I made search to get answers for those > questions but couldn't get it well > > Thanks for your time > > > On Tuesday, March 22, 2016, Erik Weathers <[email protected]> wrote: > >> Not sure what you mean by "it". The latency numbers might *never* >> stabilize depending on how your topology is written and the data it is >> processing. >> As I said, this is highly dependent on a bunch of stuff. The latency >> numbers might stabilize within minutes if the topology is trivial. Or it >> might take hours (e.g., if your topology is doing a lot of work that is >> dependent on having data pre-cached). Or it might never stabilize if the >> data flow and content is highly variable. >> >> - Erik >> >> On Mon, Mar 21, 2016 at 4:49 PM, sam mohel <[email protected]> wrote: >> >>> Thanks but how can I decide it . or what are the conditions that make >>> me know it ? >>> >>> >>> In Tuesday, March 22, 2016, Erik Weathers <[email protected]> wrote: >>> >>>> That *entirely* depends on your topology and environment. >>>> >>>> - Erik >>>> >>>> On Mon, Mar 21, 2016 at 4:27 PM, sam mohel <[email protected]> wrote: >>>> >>>>> I mean numbers which I got in complete latency and execute latency . >>>>> when I submitted topology those numbers changed every time I refreshed >>>>> page >>>>> of storm ui . when it becomes stable to measure performance of topology >>>>> >>>>> Tuesday, March 22, 2016, Justin Hopper <[email protected]> >>>>> wrote: >>>>> >>>>>> Sam, >>>>>> >>>>>> Which numbers are you speaking of? And by “get the numbers rightly”, >>>>>> what do you mean by that? >>>>>> >>>>>> Thanks, >>>>>> >>>>>> Justin >>>>>> >>>>>> On Mar 21, 2016, at 14:30, sam mohel <[email protected]> wrote: >>>>>> >>>>>> Thanks again. Last question about storm ui . when I submitted >>>>>> topology numbers that appeared changed every time I refreshed page . so >>>>>> how can I get numbers rightly ? >>>>>> >>>>>> On Monday, March 21, 2016, Justin Hopper <[email protected]> >>>>>> wrote: >>>>>> >>>>>>> Right. >>>>>>> >>>>>>> On Mar 21, 2016, at 14:16, sam mohel <[email protected]> wrote: >>>>>>> >>>>>>> Really thanks for your reply . I will try it now . but I'm asking >>>>>>> about that if I run supervisor with high core then it will give me high >>>>>>> performance for topology right ? >>>>>>> >>>>>>> On Monday, March 21, 2016, Justin Hopper <[email protected]> >>>>>>> wrote: >>>>>>> >>>>>>>> I am recommending that you that try that approach first (storm + >>>>>>>> zoo on core2 duo; super on i5). I would expect more thread contention >>>>>>>> (decrease performance) if you choose to reverse the deployment >>>>>>>> strategy. >>>>>>>> >>>>>>>> Thanks, >>>>>>>> >>>>>>>> Justin >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> On Mar 21, 2016, at 11:18, sam mohel <[email protected]> wrote: >>>>>>>> >>>>>>>> Thanks for replying . so I should run nimbus ,zookeeper on machine >>>>>>>> core2duo and supervisor on core I 5 but Are results will change if I >>>>>>>> run it >>>>>>>> in reverse . >>>>>>>> >>>>>>>> On Monday, March 21, 2016, Justin Hopper <[email protected]> >>>>>>>> wrote: >>>>>>>> >>>>>>>>> Sam, >>>>>>>>> >>>>>>>>> Nimbus and Zookeeper mostly perform management of the cluster >>>>>>>>> itself - not performing real work. Given your situation, place the >>>>>>>>> Supervisor on the high-performant system (core i5). >>>>>>>>> >>>>>>>>> Justin >>>>>>>>> >>>>>>>>> On Mar 20, 2016, at 23:15, sam mohel <[email protected]> wrote: >>>>>>>>> >>>>>>>>> Is there any help ? >>>>>>>>> >>>>>>>>> On Monday, March 21, 2016, sam mohel <[email protected]> wrote: >>>>>>>>> >>>>>>>>>> I have 2 machine one has core 2duo and other has core i5 . which >>>>>>>>>> one should I run nimbus and zookeeper on it and which one should I >>>>>>>>>> run >>>>>>>>>> supervisor ? >>>>>>>>>> >>>>>>>>>> Should worker run on machine with high core or it shouldn't ? >>>>>>>>>> >>>>>>>>>> Appreciate your help >>>>>>>>>> Thanks in advance >>>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>> >>>>>>> >>>>>> >>>> >>
