Beacon Chain Configs for 8 Validators

[terencechain]

Problem:
1.) We are starting out with 8 validators in a local test net, with 8 validators and 64 epoch length there's skipped slots:
Ex: ([]= skipped block, [#] = validator index assigned to that slot)
[[] [] [] [] [] [] [] [5] [] [] [] [] [] [] [] [7] [] [] [] [] [] [] [] [4] [] [] [] [] [] [] [] [1] [] [] [] [] [] [] [] [3] [] [] [] [] [] [] [] [2] [] [] [] [] [] [] [] [0] [] [] [] [] [] [] [] [6]]

2.)To keep things simple for the very first iteration, we don't wanna deal with skip blocks so I shortened the epoch length to 8:
[[5] [7] [4] [1] [3] [2] [0] [6]]

Question for @djrtwo, @JustinDrake and @vbuterin, what parameter numbers do you recommend for a local test net with 8 validators to start with? (SHARD_COUNT, TARGET_COMMITTEE_SIZE, and EPOCH_LENGTH)

[ralexstokes]

if you are just testing things out and scaled the epoch length by 8, try scaling everything else down by a factor of 8?

alternatively i would suggest either bumping up the number of validators, or just handling the skipped slots

[JustinDrake]

I'd do SHARD_COUNT = 1.

[prestonvanloon]

We tried

EPOCH_LENGTH = 8
SHARD_COUNT = 1
TARGET_COMMITTEE_SIZE = 1

This was causing a divide by zero issue due to an empty crosslink committee in the first epoch.
Let me step through the code and I'll add my findings to this comment.

[terencechain]

Divide by 0 is our issue and it is currently tracked by #1481. Once we fix that we can handle skip slot

[djrtwo]

This also might be an issue depending on how it's interpreted (ethereum/consensus-specs#541).

Is that issue failing due to no committee being assigned for that slot but the validator is requesting the proposer at that slot?

It is unfortunate that the most realistic early nets are at these unrealistic low numbers :/ (compared to production). The protocol generally degrades gracefully but there are certainly edge cases around helper methods failing as you are running into.

[djrtwo]

@prestonvanloon

This was causing a divide by zero issue due to an empty crosslink committee in the first epoch.

Empty as in there is not one assigned to the slot at all or because the committee didn't show up to attest?

[djrtwo]

I might even go with EPOCH_LENGTH = 4 so you get two validators per slot

[prestonvanloon]

@djrtwo we were having an issue with config overrides were not being applied.

By the way, 8 doesn’t reflect a number of validators we would deploy on a testnet. We’re just looking to see the client do stuff before standing up the infrastructure for a larger deployment.

We’ll circle back on this issue after we’ve caught up on all of the first class epoch changes. Thanks!

[djrtwo]

oh I know! I just meant these early simulations

[rauljordan]

Closed by #1565