Atomic data grain

I am having just an indecision on how to define my grain in an instance. The example is that every 5 minutes, data is polled from a device. This data may contain 4+ separate measurements each time. Overtime, these measurements will be utilized in different ways, which is not a concern. Should my Fact Table be restricted to one measurement type per fact table or should all these measurements at that specific time be added into a single fact table? An example of one poll from one device is below. I would appreciate any constructive input. Thanks in advance!!

Alton Device
Bandwidth 15436.23
Packet Loss .0001
Delay .23
Jitter .1

What do you think the grain is and why do you think the grains would be different for a device with 4 measurements vs one with 10?

What you need to be concerned about is are there different collections of measurements taking place on different equipment.  In other words are the measures significantly different or large in number across all samples?

A row should represent a collection of measurements for a particular machine at a particular point in time.  If there are a large number of very different measurements taken from different machines, it does not make sense to have one fact table with an inordinate number of measures, most of which are null.  It becomes very cumbersome to use.

If all you have are the four measurements, one fact table with four measures is fine.

If you are thinking about one measurement per row with a measurement type dimension... it is effectively the same thing as a really wide fact table, only it is even more difficult to use.

Thank you for the prompt replies. That resolved my issue. I was envisioning it to granular. Rather than looking at each poll from the device as one record, I was envisioning the grain to only obtain one metric. It did not make much sense to do it this way. I can see where not defining the grain correctly can turn into a mess. I am glad I asked. Thanks again! You provided great value!