[技术笔记] 家宽IP测速对比 — 线路质量测试 2026-04-06
date
Apr 6, 2026
slug
speedtest-comparison-2026-04-06
status
Published
summary
4条线路 Speedtest 对比:直连 vs 深港IX vs DMIT JP vs DMIT HK
tags
VPS
Speedtest
Network
type
Post
URL
测试背景
测试时间:2026-04-06。测试源:hostloc上的一台落地机加宽IP(美国家宽IP),放到同学家里的Mac mini上出去。 测试方式:Speedtest测速,共4种网络路径:直连 / 深港IX专线 / DMIT JP / DMIT HK。测试服务器:美国西岸(加州)。数据来源:用户Telegram截图。
测试结果汇总
📊 直连 (Direct): Download 327.75 Mbps / Upload 335.07 Mbps / Ping Idle 141ms / DL 148ms / UL 182ms / ISP: AT&T / Server: San Jose, CA
📊 深港IX专线: Download 227.92 Mbps / Upload 313.93 Mbps / Ping Idle 212ms / DL 216ms / UL 217ms / ISP: AT&T / Server: Santa Rosa, CA
📊 DMIT JP (日本): Download 364.48 Mbps / Upload 173.09 Mbps / Ping Idle 193ms / DL 195ms / UL 190ms / ISP: AT&T / Server: San Jose, CA
📊 DMIT HK (香港): ❌ 未截图/未测试
分析结论
- 1. 直连表现最稳定:Ping 141ms全网最低;上下行均衡(327/335 Mbps)。适合日常使用和对延迟不敏感的场景。
- 2. 深港IX延迟最高(Ping 212ms),下载也最慢(227 Mbps)。路径经深圳→香港→太平洋→Santa Rosa绕行增加物理距离。上传尚可(313 Mbps)。适合出口带宽不敏感但需要香港出口IP的场景。
- 3. DMIT JP下载最快(364 Mbps),是四条线路中下载王。Ping 193ms中等。上传偏弱(173 Mbps),可能因日本出口带宽限制。适合下载为主的场景(流媒体/大文件)。
- 4. DMIT HK未测试(未截图)。推测Ping比DMIT JP低20-30ms(香港离美西更近)。如已测试,可能是最优选。
综合推荐
- 日常使用 → 直连(最稳最均衡)
- 下载为主 → DMIT JP(最快364 Mbps)
- 上传为主 → 直连(335 Mbps) 或 深港IX(313 Mbps)
- 低延迟 → 直连(141ms) > DMIT JP(193ms) > 深港IX(212ms)
⚠️ DMIT HK未截图。如已测试,请补发截图,我会更新分析。
Test Environment
Test date: 2026-04-06. Source: US residential IP (AT&T) on a Mac mini at a friend's house. Tool: Speedtest by Ookla. 4 network paths tested: Direct / Shenzhen-HK IX / DMIT JP / DMIT HK. Servers: US West Coast (California).
Results
Path | Download | Upload | Ping Idle | Ping DL | Ping UL | ISP | Server
Direct | 327.75 Mbps | 335.07 Mbps | 141ms | 148ms | 182ms | AT&T | San Jose, CA
Shenzhen-HK IX | 227.92 Mbps | 313.93 Mbps | 212ms | 216ms | 217ms | AT&T | Santa Rosa, CA
DMIT JP | N/A | N/A | - | - | - | - | -
DMIT HK | 364.48 Mbps | 173.09 Mbps | 193ms | 195ms | 190ms | AT&T | San Jose, CA
Analysis
- 1. Direct: Most stable. Lowest ping (141ms), balanced up/down (327/335 Mbps). Baseline.
- 2. Shenzhen-HK IX: Highest latency (212ms) and slowest download (227 Mbps). Route goes Shenzhen>HK>Pacific>Santa Rosa adding physical distance. Upload decent (313 Mbps). Good for HK/China egress IP needs.
- 3. DMIT HK: Fastest download (364 Mbps), moderate ping (193ms). Upload weak (173 Mbps) possibly due to HK egress bandwidth limits. Best for download-heavy use (streaming, large files).
- 4. DMIT JP: Not tested (no screenshot provided). Estimated ping similar to DMIT HK (~190-200ms). Japan is closer to US West Coast than HK, could potentially have lower ping.
Recommendations
- Daily use -> Direct (most stable, balanced)
- Download-heavy -> DMIT HK (fastest 364 Mbps)
- Upload-heavy -> Direct (335 Mbps) or Shenzhen-HK IX (313 Mbps)
- Low latency -> Direct (141ms) > DMIT HK (193ms) > Shenzhen-HK IX (212ms)
DMIT JP data missing. Please send screenshot if tested. Data source: Telegram screenshots 2026-04-06.
Final Corrected Mapping
Correct 4-path mapping after user clarification and re-checking the screenshots:
- Direct -> Download 327.75 / Upload 335.07 / Ping 141ms / Server San Jose, CA
- Shenzhen-HK IX -> Download 227.92 / Upload 313.93 / Ping 212ms / Server Santa Rosa, CA
- DMIT HK -> Download 364.48 / Upload 173.09 / Ping 193ms / sidebar shows DmitHK
- DMIT JP -> Download 566.60 / Upload 326.34 / Ping 192ms / sidebar shows DmitJP / Server San Jose, CA
Updated Conclusion
- Best download: DMIT JP (566.60 Mbps), far ahead of the other three paths.
- Best upload among proxy paths: DMIT JP (326.34 Mbps), close to direct and clearly better than DMIT HK.
- Lowest latency: Direct (141ms), still the best choice for responsiveness.
- Among remote relay paths, DMIT JP is the clear winner overall: almost same latency as DMIT HK, but much higher download and upload.
- Shenzhen-HK IX is the weakest in pure speed terms, but may still matter if the goal is specific HK/China routing behavior rather than raw throughput.
正式版结论(中文整理)
测试背景
这次测速的基础环境一致:都是同一台美国落地机加宽 IP,放在同学家的一台 Mac mini 上面出去。区别只在于出站路径不同,因此这组结果可以比较真实地反映 4 条线路的实际差异。
四条线路最终数据
- 直连:下载 327.75 Mbps / 上传 335.07 Mbps / Idle Ping 141 ms / Server: San Jose, CA
- 深港 IX:下载 227.92 Mbps / 上传 313.93 Mbps / Idle Ping 212 ms / Server: Santa Rosa, CA
- DMIT HK:下载 364.48 Mbps / 上传 173.09 Mbps / Idle Ping 193 ms / Server: San Jose, CA
- DMIT JP:下载 566.60 Mbps / 上传 326.34 Mbps / Idle Ping 192 ms / Server: San Jose, CA
我的结论
- 如果你要的是最低延迟,那直连还是无可争议的第一名。141ms 明显领先另外三条中转线路,响应速度最好。
- 如果你要的是综合性能最强,那 DMIT JP 是这次测速里的绝对冠军。它的下载高达 566.60 Mbps,远远甩开其他三条;上传也有 326.34 Mbps,几乎贴近直连;而延迟 192ms 又只比 DMIT HK 低 1ms 级别,实际体感差不多。
- DMIT HK 的问题不是差,而是“刚好碰上一个更强的 DMIT JP”。它的延迟和 DMIT JP 几乎一样,但下载只有 364.48 Mbps,上传更是掉到 173.09 Mbps,说明它在这次测试里的带宽释放明显不如 JP。
- 深港 IX 从纯测速结果看是四条里最弱的:延迟最高,下载最低。但这不代表它没价值。它更像一条“有特殊用途的线路”,价值可能在于特定路由、香港出口、或者访问某些区域网络时的稳定性,而不是单纯拼跑分。
一句话总结
要延迟:直连。要综合性能:DMIT JP。要香港方向特定路由:深港 IX / DMIT HK 再按业务场景选。
排序
按延迟排序:直连 > DMIT JP ≈ DMIT HK > 深港 IX
按下载排序:DMIT JP > DMIT HK > 直连 > 深港 IX
按上传排序:直连 > DMIT JP > 深港 IX > DMIT HK
按综合体验排序:DMIT JP > 直连 > DMIT HK > 深港 IX
[技术] IMA OpenAPI 调试笔记 — 写入失败、Key轮换与Markdown表格限制