I think the big thing to make throughput better/actually scale would be to do bigger inserts and to only have one session inserting into a particular on-disk table at a time
I don't know of a way to do that that could possibly work out well except for in memory
@JoeObbish this is the only practical use if imoltp I've seen implemented, and it was for crazy high rate data ingestion
@jadarnel27 @Forrest think of it in this context: you have an event that lots of people want to bet on, and while it's hot there can't be any delay in them placing or changing bets, but once the event closes it only matters for historical reasons
> I'm coming back... Can't believe this show made me feel so much emotion. Watching for the second time won't feel the same but this show was absolutely inspiring.
Basically we have a listener that points to 2 servers. Server A is the primary DB server and server B is the secondary. Some of the databases on server A are not part of the AG and therefore if the listener points to the secondary those DB's are not available using the listener.
Which is why I want them to stop using the connection for the listener and instead just connect to server A.
That's always what they were supposed to be using but they don't want to use that for... Reasons. They don't want to be on the AG they just want to keep using that listener because it's "what everyone else does"...
So basically this weekend they lost connection and because I resolved the issue, it must be a DATABASE PROBLEM.
If not all employees are shift-based, but rather a small number are, then you can add MAXDOP hint to further encourage the optimizer to use an optimized bitmap filter. Consider removing the SELECT * and only retrieving what you really need, too, as you're technically returning duplicate data and therefore increasing network I/O consumption unnecessarily. I'm sure you already know that... Your current tables are small but these issues will show up over time. Best to code it right the first time. — John Zabroski17 mins ago
I figured out the answer to my question - sort of.
When you specify MAXDOP(8), the optimized bitmap filter optimization will create 8 hash buckets. The exact spread of integers across those hash buckets is probably Pearson hashing. I am curious if anyone in the comments knows the exact hashing...