大网数据广告活动
服务器运维问答

Disk iops vs latency

· 52CE 我爱测

I am trying to understand the diference between iops and latency, this is my question: If an hypothetical disk had 2 io per second, then its latency would be 0.5 seconds?

问题原文

原帖链接 · Javier Cintrón · CC BY-SA 4.0

I am trying to understand the diference between iops and latency, this is my question:

If an hypothetical disk had 2 io per second, then its latency would be 0.5 seconds?

已采纳解答

原帖链接 · poige · CC BY-SA 4.0

I am trying to understand the diference between iops and latency, this is my question:
If an hypothetical disk had 2 io per second, then its latency would be 0.5 seconds?

  • IOPS is the number of I/O operations completed per-second
  • Latency implies there's a delay before IO completes not only due its volume (amount of data is being sent) but even when it's absolute minimal data requested. Thus it's kinda penalty each separate IO operation incurs except the transfer time itself — which is bandwidth dependent.

Let's get back to:

If an hypothetical disk had 2 io per second, then its latency would be 0.5 seconds?

Hopefully, now it's clear that the answer here is "it could be". Because if each I/O was lengthy (latency 0.1 + sending data 0.4 sec) it's obviously them 2 would need totally 1 second to complete but by definition (formula) latency was just 0.1 sec. OTOH, if you know that no matter how minimal was the data transfer, but only 2 IOPS completed, then, yeah, it's latency 0.5 sec.

P. S. Similar things happen in Networking domain, BTW. You run ping with some minimal data length possible and get estimation of channel's latency, then you increase its length or download big enough file with curl and figure out channel's bandwidth.

摘自公开归档,保留原文语言,版式经整理,外部媒体与邮箱地址已移除。

资料来源与适用范围

资料来源
Server Fault
适用范围
以原帖系统、软件版本与发布日期为准(2022-01-24)
内容语言
中文;命令、路径和错误标识保留技术原文
本地收录日期
2026-11-15
授权说明
CC BY-SA 4.0

正文与所需资源由本站本地存储。查看来源及更新记录 ↗

请按适用版本和实际日志确认原因;本站没有对每一种系统或设备逐项执行文中的操作。