How big are the jobs being done? I remember reading a statistic[1] that is a couple of years old that 50% of the big data jobs were less than 10 GB. 90% was less than 100 GB and only something like 1% was bigger than a TB. You can fit multiple terabytes on a single machine. I wonder if anyone has new statistics about this.
[1]: http://www.msr-waypoint.com/pubs/204499/a20-appuswamy.pdf
I have good context on this from seeing the analytics warehouse for a very popular social network. Even with a multi-terabyte hot database, the analytics dataset fit on a single 2U server full of drives. I have not seen a deployment of Hadoop yet that makes sense. Usually I see things like a 2 or 3 TB log spread across 60 nodes and a job to count entries matching a regular expression (because Hadoop becomes the hammer hunting for nails), which can easily be done in a VM on a Mac in minutes.
That was the thrust of the original post, and I agree wholeheartedly.
Yes, I agree that it's not an uncommon situation. But regardless whether legitimately big deployments are common or not, I never really appreciated some negligence with performance in using Java for such things. Just because it scales, it doesn't mean that overhead should be ignored, because the cost of the overhead scales too.
So going back to my original point, having such kind of systems made with Rust would be really much better.
I don't like the 90% argument. 90% of apps are just dumb frontends to a MySQL database, so you don't need Rust, or Spark. As you can see, that doesn't tell you anything meaningful about the other 10%.