Swap your aggregations. Put the x-axis/date split first, then the Split Bars/application second.
Elasticsearch slices and dices you data up in the order that you've specified the aggregations. So, first it has to group all records by application, and then it breaks them all apart by date. It's cheaper to group by date first.
Also, in the split bars aggregation, in the advanced section, if you've moved the size too high, you're going to get back way too many records to render. The default is 5.
Generally speaking, you shouldn't need the "long tail" of your data, so you can ignore records that are common. For example, if you expect load times of 300ms or less, you don't need to see that all systems are operating within spec, you really only care about systems that out of spec, so you can ignore the "correct" values. This also prevents information overload, where records you don't care about start to blend in with and hide records you need to know about. So what you need to figure out is how to show just the information that matters, which will make the visualizations (and dashboards containing them) more useful, and have the added benefit of being much faster.