Use larger instances with more CPUs

[j1010001]

trafficstars

• [x] Request m3-ultramem-128 and m2-megamem-416 instances
• [ ] Test which instance performs better
• [ ] Keep the instance that performs best, remove the other one

[j1010001]

Slack thread: https://dapperlabs.slack.com/archives/C015G65HR2P/p1714419791630609

[j1010001]

m3-megamem-128 is the fastest instance so far. Will also use C3D next.

[fxamacker]

• [x] Request m3-ultramem-128 and m2-megamem-416 instances

m3-megamem-128 is the fastest instance so far. Will also use C3D next.

@j1010001 Did you mean m3-megamem-128 (1.9TB RAM) or m3-ultramem-128 (3.9TB RAM)?

atree inlining migration requires >2TB RAM (mostly because it uses large non-inlined state as input).

[j1010001]

• [x] Request m3-ultramem-128 and m2-megamem-416 instances

m3-megamem-128 is the fastest instance so far. Will also use C3D next.

@j1010001 Did you mean m3-megamem-128 (1.9TB RAM) or m3-ultramem-128 (3.9TB RAM)?

atree inlining migration requires >2TB RAM (mostly because it uses large non-inlined state as input).

You are right @fxamacker , I meant m3-ultramem-128.

[j1010001]

c3d-highmem-360 is marginally better, but not significant.