cse压测时cup占用异常

场景：
os:
cpu:
all usage: 76.98% all idle: 23.02% process: 602.65%
net:
send(Bps) recv(Bps) send(pps) recv(pps) interface
35 35 0 0 lo
19.448M 14.379M 5.721K 11.214K eth0
vertx:
instances:
name eventLoopContext-created
registry 0
registry-watch 0
transport 0
transport:
server.endpoints:
connectCount disconnectCount rejectByLimit connections requests latency send(Bps) receive(Bps) listen
53 53 0 50 23487 17 107.930M 717.799K 10.33.205.58:8250
threadPool:
coreSize maxThreads poolSize currentBusy rejected queueSize taskCount taskFinished name
16 64 26 2 NaN 0 194.1 194.0 cse.executor.groupThreadPool-group1
16 64 26 0 NaN 0 194.9 195.0 cse.executor.groupThreadPool-group0
producer:
simple:
status tps latency operation
rest.200 0.1 0.240/0.255 NormalContentService.normalContentCheck.getMicroServerHeartBeat
391.3 6.354/554.521 NormalContentService.BrowserNormalContent.queryNormalDocs
391.4 6.353/554.521 (summary)
details:
rest.200:
NormalContentService.normalContentCheck.getMicroServerHeartBeat:
prepare: 0.017/0.023 queue : 0.028/0.030 filtersReq : 0.025/0.026 handlersReq: 0.106/0.117
execute: 0.028/0.034 handlersResp: 0.001/0.001 filtersResp: 0.009/0.010 sendResp : 0.012/0.013
NormalContentService.BrowserNormalContent.queryNormalDocs:
prepare: 0.019/106.530 queue : 0.605/113.083 filtersReq : 0.033/97.026 handlersReq: 0.104/106.444
execute: 4.325/548.923 handlersResp: 0.001/0.051 filtersResp: 1.247/553.132 sendResp : 0.018/79.599

**通过jprofile分析
cpu占用 方法
75% io/netty/handler/codec/compression/JdkZlibEncoder.deflate
12% org/apache/servicecomb/common/rest/codec/produce/ProduceJsonProcessor.doEncodeResponse
10% 业务自身
请帮看下有没有什么分析优化思路**

复现 CSE 负载测试场景并检查 JProfiler 结果，从 io/netty/handler/codec/compression/JdkZlibEncoder.deflate 和 org/apache/servicecomb/common/rest/codec/produce/ProduceJsonProcessor.doEncodeResponse 开始。比较调查的任何变更前后的 CPU 使用率和响应延迟；完成时应确定一个经过验证的原因和一个优化方向。

由索引模型根据 Issue 内容生成。