Turns Out AI Is Not Good at Database Transaction Scheduling

#1

Post by accheng » Thu, Nov 13, 2025, 8:50 PM UTC

Turns Out AI Is Not Good at Database Transaction Scheduling
adrs-ucb.notion.site

Re: Turns Out AI Is Not Good at Database Transaction Scheduling

#2

Post by zeristor » Thu, Nov 13, 2025, 9:17 PM UTC

We modelled using Redshift scheduling automation, that suggested a welcome 20% improvement.

We applied it and it run 20% slower.

As it runs it is supposed to learn how to optimise scheduling. Not sure how it gets over the burst at month, quarter, and year end.

Re: Turns Out AI Is Not Good at Database Transaction Scheduling

#3

Post by accheng » Thu, Nov 13, 2025, 9:38 PM UTC

zeristor wrote: post #2

We modelled using Redshift scheduling automation, that suggested a welcome 20% improvement. We applied it and it run 20% slower. As it runs it is supposed to learn how to optimise scheduling. Not sure how it gets over the burst at month, quarter, and year end.

What did you use to try to optimize scheduling?

Re: Turns Out AI Is Not Good at Database Transaction Scheduling

#4

Post by qat321 » Thu, Nov 13, 2025, 9:41 PM UTC

turns out DB transactions are a hard problem...

Re: Turns Out AI Is Not Good at Database Transaction Scheduling

#5

Post by taqpos » Thu, Nov 13, 2025, 9:42 PM UTC

I wonder what the optimal for these workloads are?

Re: Turns Out AI Is Not Good at Database Transaction Scheduling

#6

Post by pos456 » Thu, Nov 13, 2025, 9:44 PM UTC

For the offline problem, AI found a new solution that was 34% faster. Interesting results!

Re: Turns Out AI Is Not Good at Database Transaction Scheduling

#7

Post by gfhsad » Thu, Nov 13, 2025, 9:45 PM UTC

What about other use cases

Re: Turns Out AI Is Not Good at Database Transaction Scheduling

#8

Post by kap901 » Thu, Nov 13, 2025, 9:46 PM UTC

Seems like AI hype

Re: Turns Out AI Is Not Good at Database Transaction Scheduling

#9

Post by pakt1 » Thu, Nov 13, 2025, 9:48 PM UTC

Transactions seem like a hard problem for scheduling, wonder how the authors got this to work in the original paper?

Re: Turns Out AI Is Not Good at Database Transaction Scheduling

#10

Post by accheng » Thu, Nov 13, 2025, 9:50 PM UTC

pakt1 wrote: post #9

Transactions seem like a hard problem for scheduling, wonder how the authors got this to work in the original paper?

As described in our VLDB paper (https://www.vldb.org/pvldb/vol17/p2694-cheng.pdf), we had to implement a number of optimizations and integrate scheduling with CC to get good performance.