Hi @Joe Harvy, Because LLAP uses its own AM (application managers) based on query fragments instead of Tez AM which are based on sessions and are brought down when the session is done, LLAP is capable of higher concurrency than Tez alone. With that being said, you'd probably not want more than 100 concurrent users on LLAP. Keep in mind though that is 100 users executing a query at the same time. This scenario would most likely mean you have 1,000's of users on the system which is rare for most BI workloads. Druid has the same limitations since it is tightly integrated with Hive.
If you must have 100's or 1,000's of concurrent users then look to something like IBM's BigSQL.