Based on the press, we know how this is going to turn out already, now don’t we? You go to a Friday the 13th movie, you know some campers are going to bite it. But what the hell, lets play anyway.
Here’s what I did:
- Bounced Tableau Server on my “big” v8 machine (an 8 core beaut with 64 GB of RAM)
- Executed a sampling of reports: some very simple, some complex, some uber-complex
- Recorded execution time of each
- Upgraded to v9
- Ran reports again
First, here are the vizzes in my workload – most of them will look familiar as they have been Tableau samples at some point:
The vizzes that you see “crossed out” are sheets which are part of a Dashboard called “Euronext” – A 200M+ data source I used for the semi-complex load. I never actually tested all of its worksheets individually, just the dashboard…but I’m too lazy to change the screenshot. Sorry.
The grey vizzes come to me by way of customers and are my “really complex” workload. Each of dashboards contains many vizzes which utilize blending against 3-4M+ row data sources. There are all sorts of fun, complex calcs in each one, including use of the dreaded CountD. Plenty of table calcs, too. Some render many marks, some render many marks on top of maps. These take time to execute.
Now, the numbers:
Across the entire sample, I saw a 61% increase in performance excluding some ridiculously awesome results that were just unfair to poor old v8. I could have easily made these numbers “better” if I had chosen more vizzes which take a while to render, but I didn’t. Also, keep in mind these vizzes all run off extracts, so some of the “goodness” that we might normally get in 9 (like Query Fusion) is off the table.
What are the ridiculously awesome results I am hiding by default? Euronext. These puppies made it into the external cache of v9 and remember that the external cache sticks around even after you cycle Tableau Server. Because these were already cached, They rendered 2000%+ more quickly on v9. I didn’t want to embarrass my old friend v8, so I omitted them. If you add them BACK in with the filter, you’ll see an 86% overall performance increase on the workload. Like I said, this was like taking candy from a baby so I tried to be kind.
Nice boost in performance for a 20 minute upgrade, eh?
Finally, guess what this is:
If you guessed “It’s over 215K dashboard executions as Russell did v8 vs. v9 load testing on the same workload”, you’d be right. But we’ll talk about that another day.