adjust tpu coalesce channel size to 250K

[lijunwangs]

Problem

Based on further testing of TPU coalescer channel size on the receive performance. Adjust the size to reduce memory usage.

Summary of Changes

Found the channel size with 250K has similar performance with the size of 1 million (2.8 M/5s). When the size drops to 100K, it drops to about 2.18 M/5s

Fixes #

[alexpyattaev]

LGTM but please document how you have measured the performance (and on which hardware) so we can reproduce/backtrack later.

[lijunwangs]

Using bench-vote tool

Run server

./target/release/solana-bench-vote --use-connection-cache --num-recv-sockets 32 --num-producers 1 --server-only --server-address 0.0.0.0:11228 --verbose --identity ~/validator-keypair.json --use-quic true

Run client:

./target/release/solana-bench-vote --num-recv-sockets 1 --num-producers 32 --client-only --server-address 145.40.90.189:11228 --use-quic true --identity ~/validator-keypair.json

Then monitor the new_streams in the metrics bench_vote_metrics.

Hardware:

*-core
description: Motherboard
physical id: 0
*-memory
description: System memory
physical id: 0
size: 512GiB
*-cpu
product: AMD EPYC 7502P 32-Core Processor
vendor: Advanced Micro Devices [AMD]
physical id: 1
bus info: cpu@0
version: 23.49.0
width: 64 bits

I was using the branch https://github.com/lijunwangs/solana/tree/bench_vote_quic_mcep.2 which has some tweaks on the client side to maximize performance for this exercise.