The way we selected our workbooks for tabjolt is that we decided on how many workbooks to test out and arrived at the number 20. And then we came up with the list by figuring out which were the most common used dashboards, which were the most critical to the business, and which ones have been previously problematic.
Hi Jeff, thanks for the reply, and sorry for getting back late, I've been troubleshooting some Tabjolt errors and finally sorted them out.
Is there a way to determine a 'real' capacity for the server? Such as determining how much load your hardware can handle after an upgrade? I was hoping to get a simple number for our current vs future hardware. For example "with an average load time of 2s we can currently support 100 concurrent active users where as with the new hardware we can support 150". Using a very select group of workbooks doesn't reflect closely on the real world. Would we be able to extrapolate the findings in Tabjolt to reflect the number of users we can handle better?
In our case, the 20 workbooks that we selected represent about 10% of the workbook inventory in the primary site, but represent about 60 to 70% of the actual usage. So if you can find such a correlation of heavily hit workbooks that takes up the majority of usage, then this is one way of handling it.
We have about 2000 workbooks so it wont be nearly as accurate but I'll at least be able to pick the most viewed dashboards of late and use that with the Tabjolt results to see roughly what we can handle.
Edit: Were you able to calculate that 60%-70% usage?
Edit2: I looked back to this white paper that mentions creating a weighted average from Average Response time X number of requests
I'll bring up the specific fields when I get a chance and update here.
I think I'll be using the list created from above, that way we'lll be able to test the workbooks that create the most load. Though it'd still be unrealistic to test a large number of them
Is there a way to test/simulate extract loads as the same time as workbooks? Realistically both would be running simultaneously backgrounders and VizQL processes.