Tracking parameters for knuu e2e benchmark tests under network latency

[staheri14]

Problem

Part of the Testground sanity test is to run benchmark tests in Knuu under network latency. This issue is to track all the desired test scenario, and relevant parameters related to these experiments, ensuring consistency across the two backends.

Acceptance Criteria

The following aspects need to be agreed upon/investigated:

• Amount of Added Latency: fixed 70 ms
  • As per a sync conversation with @evan-forbes, the network latency used in the Testground experiment has been set to a fixed 70ms. The reason for not using a random value is to make it easier to debug and reason about the observations from the experiment. Hence, the same value should be replicated in Knuu.
• Level of Latency Application: TBD
  • There is some ambiguity around where and at which level this delay should be added (e.g., per TCP packet or per reactor message). This part needs further investigation.
  • Test Scenario: TBD
  • Previously, for other benchmark tests, we ran network tests with 2 validators and 50 validators, and for max block sizes of 8, 32, and 64 MiB. We can replicate the same for tests with network latency, or just focus on 50 validators and block size of 64 MiB. cc: @evan-forbes

[staheri14]

Echoing some of the past conversations with the DevOps regarding how latency is managed in knuu

• To apply latency and jitter TC is used behind the scene. This might help: https://www.man7.org/linux/man-pages/man8/tc.8.html
• This one might help to understand better how latency & jitter are applied: https://github.com/celestiaorg/bittwister/blob/main/xdp/latency/latency.go#L75

cc: @evan-forbes You also mentioned that Testground utilizes TC under the hood, so I guess, it implies that the way the latency is applied in both platforms is similar. I will do further research to confirm this.

[staheri14]

I believe for the other parameters, we can mirror the settings we have used in the benchmark tests w.o. latency.

Parameter	Value
WAL	true
cpu	8
PeerGossipSleepDuration	100ms
memory	12GiB
latency	70ms
random_latency	false
bandwidth	1GB
topology	seed
pex	true
timeout_propose	80s
timeout_commit	11s
per_peer_bandwidth	5MBs
blob_sequences	3
blob_sizes	200000
blobs_per_sequence	6
inbonund_peer_count	40
outbound_peer_count	10
gov_max_square_size	512
max_block_bytes	7_800_000 (this varies depending on the targeted square size in the experiment)
mempool	v1
broadcast_txs	true
proposer_gossip_fix	false
send_prio	ratio
receive_prio	fifo
compact block prop	false
BlockPartSize	64KB
Mempool.MaxTxsBytes	1_000_000_000
Mempool.MaxTxBytes	100_000_000
Mempool.TTLNumBlocks	100
Mempool.TTLDuration	40 * time.Minute