Thread by @DavidDaly44

Question for my #performance folks: Any new ideas or insights on getting reproducible performance in the cloud (and AWS in particular)? Specifically on configuring a cloud instance to minimize performance variability /1

My team has spent a lot of effort controlling noise in the past, disabling hyperthreading, using numactl, disabling frequency scaling and sleep states, using placement groups, etc. /2
engineering.mongodb.com/post/reducing-…

I would love to learn more knobs to make things better (lower variation). In particular, we’ve found newer processors and cloud instances to have more performance variation than older ones, and we need to deal with it. /3

This is along the lines of what already is documented by AWS. I'm aware of those options, but I have some limits on their use as they often require using the largest instance type, and those sizes (and costs) keep getting larger. /4 docs.aws.amazon.com/AWSEC2/latest/…

While I'm here, I am aware of a lot of great work on performance variability in the cloud. I'll link some of it in, starting with "Taming Performance Variability" /5 usenix.org/system/files/o…