tomlurge / visionion Goto Github PK

When we divided numbers of users per day by 24 in order to get the number of users per hour, this was a mistake. We should have written the very same number of users in the hourly data.

Here's why: we calculate the number of users per day by dividing the total number of consensus requests by 10, assuming that every continuously connected client makes 10 consensus requests per day. The result is the theoretical average number of clients that are connected at any given time on that day. There could have been 10% fewer users in the morning and 10% more users in the afternoon, but we won't find out. This number of users is the same when considering the whole day or considering a single hour of it. That's why we need to undo the step where we divide by 24.

Similarly, when we want to know the number of users per week, we can only calculate the average of the 7 days in a week. We wouldn't want to sum up these 7 numbers and state that this is the number of users per week, because what would that tell us?

To give another example, if we knew that there were 480 relays on Linux on a given day, and we wanted to know that number per hour, we wouldn't divide by 24 and say that there were 20 relays at 00:00. The reason is that 480 is already a mean value computed over an entire day, just like the user number per day is a mean value calculated over the day.

I can prepare a patch to the Java importer if you agree with this change.