[{"data":1,"prerenderedAt":211},["ShallowReactive",2],{"blog-\u002Fblog\u002Fji-chang-ce-su-gong-ju-tui-jian":3},{"id":4,"title":5,"author":6,"body":7,"category":195,"date":196,"description":197,"extension":198,"image":199,"meta":200,"navigation":201,"path":202,"seo":203,"stem":204,"tags":205,"__hash__":210},"blog\u002Fblog\u002Fji-chang-ce-su-gong-ju-tui-jian.md","机场测速工具怎么选：客户端检测与 iperf3 分别测什么","品牌图鉴编辑部",{"type":8,"value":9,"toc":185},"minimark",[10,17,22,25,29,32,50,53,57,60,120,123,126,130,133,137,155,158,164,170,176,179],[11,12,13],"p",{},[14,15,16],"strong",{},"客户端的测速按钮显示的是握手往返时间，只代表当下一刻；iperf3 测的是 TCP 或 UDP 的吞吐量。要判断速度，请拿真实下载来验证，看几分钟内的持续速率，别只看峰值。",[18,19,21],"h2",{"id":20},"客户端的测速按钮测的是什么","客户端的测速按钮测的是什么？",[11,23,24],{},"按钮给出的毫秒数，衡量的是建立连接时一来一回花掉的时间，反映的只是按下去的那一瞬间。 想让结果可信，就在早中晚各点几次。 带宽高低则得靠真实下载来确认，盯住几分钟内速率的起伏，别被偶然冲高的峰值带偏。",[18,26,28],{"id":27},"内核是怎么做健康检查的","内核是怎么做健康检查的？",[11,30,31],{},"两个内核都能让策略组按固定节奏自动探测节点，只是参数叫法和默认值各有不同。",[33,34,35,44],"ul",{},[36,37,38,39,43],"li",{},"在 mihomo 一侧，interval 以秒计，填非 0 的数就开启周期检测；tolerance 以毫秒计，决定延迟差多大才换节点；探测地址在文档示例里用的是 gstatic 提供的 ",[40,41,42],"code",{},"generate_204","。",[36,45,46,47,49],{},"在 sing-box 的 urltest 一侧，interval 默认 3 分钟，tolerance 默认 50，url 留空时同样落到 gstatic 的 ",[40,48,42],{},"，另有 idle_timeout，默认 30 分钟。",[11,51,52],{},"mihomo 里还有 lazy 参数，缺省为 true：当前没有被选用的策略组会跳过探测，省去无谓的请求。timeout 以毫秒计。max-failed-times 缺省为 5，累计失败达到这个数，会马上补做一次健康检查。 订阅提供者（proxy-provider）也可以各自带一段 health-check 配置，字段有 enable、url、interval、timeout、lazy 和 expected-status。 sing-box 方面，interrupt_exist_connections 缺省为 false，它管的是所选出站变动时，已建立的连接要不要被掐断。",[18,54,56],{"id":55},"iperf3-是做什么的","iperf3 是做什么的？",[11,58,59],{},"iperf3 文档把它描述为测 TCP 和 UDP 网络吞吐量的工具。常用的几个选项如下。",[61,62,63,76],"table",{},[64,65,66],"thead",{},[67,68,69,73],"tr",{},[70,71,72],"th",{},"选项",[70,74,75],{},"含义",[77,78,79,88,96,104,112],"tbody",{},[67,80,81,85],{},[82,83,84],"td",{},"-u",[82,86,87],{},"用 UDP",[67,89,90,93],{},[82,91,92],{},"-R",[82,94,95],{},"反向模式，服务端发、客户端收",[67,97,98,101],{},[82,99,100],{},"-P",[82,102,103],{},"并行连接数，默认 1",[67,105,106,109],{},[82,107,108],{},"-O",[82,110,111],{},"忽略测试开头若干秒，避开 TCP 慢启动",[67,113,114,117],{},[82,115,116],{},"-i",[82,118,119],{},"周期性输出带宽、抖动和丢包报告",[11,121,122],{},"iperf3 具体怎么计算抖动和丢包，所取文档里没有说明。",[11,124,125],{},"测速结果受本地带宽、测试服务器位置和时段影响，同一个节点换个工具，数值可能差异明显。",[18,127,129],{"id":128},"哪款工具最合适","哪款工具最合适？",[11,131,132],{},"没有官方文档推荐某一款机场测速工具，这一点没法给出答案。 能说清的是方法：先分清你要看延迟还是吞吐，再选对应的检测方式，再用实际下载交叉验证一次。 站内的做法是同一节点多次、多时段测，不拿单次数字下判断。",[18,134,136],{"id":135},"官方文档怎么写","官方文档怎么写？",[11,138,139,140,145,146,150,151,43],{},"上面的默认值和选项读取于 2026-11-01，页面没有标日期，版本变化后可能不同。 延迟的概念见 ",[141,142,144],"a",{"href":143},"\u002Fblog\u002Fjichang-jiedian-zenme-kan\u002F","节点信息怎样读","，晚高峰变慢的情况见 ",[141,147,149],{"href":148},"\u002Fblog\u002Fjichang-wan-gaofeng-zhenduan-2026\u002F","晚高峰变慢的诊断","，相关名词见 ",[141,152,154],{"href":153},"\u002Fglossary\u002F","术语表",[18,156,157],{"id":157},"常见问题",[11,159,160,163],{},[14,161,162],{},"Q：客户端测出来延迟低，就说明速度快吗？","\n不能。延迟反映握手往返时间，速度要用实际下载验证。",[11,165,166,169],{},[14,167,168],{},"Q：iperf3 的 -R 是什么意思？","\n是反向模式，由服务端发送、客户端接收。",[11,171,172,175],{},[14,173,174],{},"Q：同一个节点换工具为什么数值不同？","\n站内的说法是测速受本地带宽、测试服务器位置和时段影响。",[177,178],"hr",{},[11,180,181],{},[182,183,184],"em",{},"依据:官方文档与通用说明 · 最近核对 2026-11-01",{"title":186,"searchDepth":187,"depth":187,"links":188},"",2,[189,190,191,192,193,194],{"id":20,"depth":187,"text":21},{"id":27,"depth":187,"text":28},{"id":55,"depth":187,"text":56},{"id":128,"depth":187,"text":129},{"id":135,"depth":187,"text":136},{"id":157,"depth":187,"text":157},"硬核网络","2026-11-08","客户端里的测速按钮测的是握手往返时间，iperf3 测的是吞吐量，两者不能互相替代。这里讲清各自测什么、有哪些相关设置，以及测速结果为什么会因工具而异。","md","\u002Fimages\u002Fblog\u002Fdefault.jpg",{},true,"\u002Fblog\u002Fji-chang-ce-su-gong-ju-tui-jian",{"title":5,"description":197},"blog\u002Fji-chang-ce-su-gong-ju-tui-jian",[206,207,208,209],"机场测速","测速工具","iperf3","延迟","5pwLWTIP6oR9CN8_td_MkQTlBcstu7R4YK0Xr4xOpRI",1794113848317]