This is way better in master (I don’t recommend use master for now), probably the work on it will be released in v1.1
https://github.com/dgraph-io/dgraph/commit/fe07e3e8a859f94411af7869e5ab1c72450e0a20
lru_mb today no matters that much. Is just a cache option. And for now is disabled by default until we finish our own cache system.
You didn’t said what type of storage you have. So you’re using --badger.vlog=disk - This param for HDD isn’t that good. Cuz HDD has low IOPS.
Okay, disable it and test again. Also decrease the load a bit, give it more air to breath. Or add more machines (real machines) each one with a Node and create a load balancing, eg: create 3 Alpha Nodes and give each of them 166.66 loads (of 500 connections that you mentioned).