喔。我多试试各种情况吧。感觉阿里云上部署的server没有这种问题，不知道是不是阿里云公网网速慢，传过去需要时间很长。我现在是内网测试，内网的网速 很大，一下子 数据全部发了过去。（猜测）
那我把server端的CPU改成64核心也试试看。
公网的话，很可能是网络限制，你先逐渐增量的增加连接数。有本地环境的进行对比更好。
看起来不是CPU的问题，CPU已经加到了64核心，看起来占用率不高，然后订阅数10000发过去，mqtt一分钟不到 就挂了，得systemctl start emqx开启服务。
提示这个，跟测试工具emqtt_bench是不是有关系？
client(2025): EXIT for {shutdown,tcp_closed}
client(2004): EXIT for {shutdown,tcp_closed}
=CRASH REPORT==== 25-Jun-2021::17:23:48.203127 ===
crasher:
pid: <0.631.0>
registered_name: []
exception exit: {noproc,
{gen_statem,call,
[<0.632.0>,
{publish,
{mqtt_msg,0,false,false,undefined,
<<"test">>,#{},<<"aaaaaaaaaaaaaaaa">>}},
infinity]}}
in function gen:do_call/4 (gen.erl, line 218)
in call from gen_statem:call_dirty/4 (gen_statem.erl, line 671)
in call from emqtt_bench:publish/2 (/root/emqtt-bench-master/src/emqtt_bench.erl, line 470)
in call from emqtt_bench:loop/5 (/root/emqtt-bench-master/src/emqtt_bench.erl, line 410)
ancestors: []
message_queue_len: 1
messages: [{'EXIT',<0.632.0>,{shutdown,tcp_closed}}]
links: [<0.115.0>]
dictionary: [{rand_seed,{#{jump => #Fun<rand.3.47293030>,
max => 288230376151711743,
next => #Fun<rand.5.47293030>,type => exsplus},
[219000802233233608|107032127440428347]}}]
trap_exit: true
status: running
heap_size: 610
stack_size: 26
reductions: 354091
neighbours:
sent(32000): total=40359, rate=1685(msg/sec)
pub(32000): total=40359, rate=1680(msg/sec)
这个是你的内网还是公网，如果公网，测试的时候，你长ping 大包一下，看看丢包如何
