Additional experiments for WWW25

Exp 1: Testing Wukong with different $\Delta$ parameters

1. Pic1

Setting:

• node number: $n=10$
• favorable situation: $f=0$
• $\Delta$: 0.5s and 1s

Results:

Wukong with $\Delta$ as 1s outperforms that with 0.5s in the favorablt situation.

2. Pic2

Setting:

• node number: $n=10$
• unfavorable situation: $f=3$
• $\Delta$: 1s and 4s

Results:

Wukong with $\Delta$ as 1s outperforms that with 4s in the unfavorable situation.

Exp 2: Testing Wukong with random message delay

Setting:

• node number: $n=10$
• favorable situation: $f=0$
• $\Delta$: 1s
• Random message delay: Each message has a 10% probability of incurring a 50-millisecond delay.

Results:

• Pic1: Latency v.s.batch size
• Pic2: Throughput v.s. batch size
• Pic3: Latency v.s. throughput

The three pictures show that even with random message delays, Wukong still outperforms the others by a significant margin.

Exp 3: Adding a new baseline: LightDAG

Due to time constraints, we have currently added only one new baseline, LightDAG. We obtained the source code for LightDAG from its authors and conducted the corresponding experiments, as shown below. We will consider adding one or two more baselines in the future.

We specifically choose LightDAG1 from [IPDPS'24] LightDAG: A Low-latency DAG-based BFT Consensus through Lightweight Broadcast 10.1109/IPDPS57955.2024.00093 as our additional baseline.

Setting:

• node number: $n=10$
• favorable situation: $f=0$
• $\Delta$: 1s

Results:

• Pic1: Latency v.s.batch size
• Pic2: Throughput v.s. batch size
• Pic3: Latency v.s. throughput

Since LightDAG overlooks the latency optimization of non-leader blocks, it also performs worse than Wukong.