I need advice on what is the appropriate approach for communicating thorough a graph the low and high work loads of my web server. To be able to distribute the computers resources at the appropriate times of heavy and low workloads. I have the number of HTTPS requests over a time period. Additionally I have the data over the same period that shows the number of unique users that are signed in longer then half hour. These users are the users that matter and use the most resources and I want to distribute my memory and CPU accordingly. The half hour doesn't have to be continuous but at least a total of 1/2 hour or more within the hour to be considered an user. I was thinking of doing a graph that shows the rate, HTTPS request per the number of users over a interval (i.e per hour). Does this make sense? Does it also make sense to do 95th percentile of HTTPS requests. Not sure how you do that though. If I look at a period of one week at a interval of one hour, for example 25th to 31st, would my data be more accurate if go back as far as a year and compute the 95 percentile of HTTPS requests? Then I would divide that 95 percentile for that interval of hour by current users of the actual hour that I was looking at?
Example: I have 100 request between 2PM and 3PM and during that hour I have 5 users. 100 /5 = 20 HTTPS per user during 2 PM and 3 PM ...that would point 1 and then I would do the same for point 2 but hour later from 3 PM to 4 PM 100 HTTPS request with 20 users 100 /20 = 5 HTTPS request per user and I continue to do this until what every period I defined, lets say a week.