program-runtime: double program cache size

[alessandrod]

The cache is currently getting thrashed and programs are getting reloaded pretty much at every single slot. Double the cache size, which makes reloading happen only due to random eviction sometimes picking a popular entry.

The JIT code size with the new cache size is about 800MB.

This change reduces JIT time 15x.

Left is with PR, right is master.

[mergify]

Backports to the beta branch are to be avoided unless absolutely necessary for fixing bugs, security issues, and perf regressions. Changes intended for backport should be structured such that a minimum effective diff can be committed separately from any refactoring, plumbing, cleanup, etc that are not strictly necessary to achieve the goal. Any of the latter should go only into master and ride the normal stabilization schedule. Exceptions include CI/metrics changes, CLI improvements and documentation updates on a case by case basis.

[bw-solana]

LGTM.

800MB seems like a small price to pay. I'm seeing ~100ms per slot spent filtering/replenishing program cache over the last 7 days on mainnet.

[alessandrod]

I'm seeing ~100ms per slot spent filtering/replenishing program cache over the last 7 days on mainnet.

Yep one odd thing I noticed is that our devboxes and canaries have about 2x worse mean program cache perf than mnb. I don't know if it's because we've got potatoes compared to mnb nodes, or if we have an actual regression in master

[Lichtso]

or if we have an actual regression in master

Can you run a v2.0 node on your devbox for comparison?

[bw-solana]

I don't know if it's because we've got potatoes compared to mnb nodes, or if we have an actual regression in master

Sample size of 1, but mds3Df1ieBonG2qS8ZoKTqshq5MgTUNfZgc78cjiCdq is running v2.1.1 on mainnet and looks significantly worse than cluster mean

[mergify]

Backports to the stable branch are to be avoided unless absolutely necessary for fixing bugs, security issues, and perf regressions. Changes intended for backport should be structured such that a minimum effective diff can be committed separately from any refactoring, plumbing, cleanup, etc that are not strictly necessary to achieve the goal. Any of the latter should go only into master and ride the normal stabilization schedule.