Servers, Storage and Virtualization · #47

Online Storage Latency and Queue Depth Calculator

Uses Little's law to relate outstanding I/O, IOPS and latency.

  • Data is processed in your browser and not sent to a server

Tool

Completed I/O per second.

Milliseconds; esxtop DAVG/GAVG or Windows 'Avg. Disk sec/Transfer' × 1000.

Average outstanding I/O.

What does this tool do?

When a VM is 'slow', the storage view shows high latency, but latency is directly linked to the number of concurrent I/Os (queue depth). Little's law (L = λ × W) lets you derive any one of the three from the other two and shows whether measurements are consistent.

How to use it

  1. Enter two of IOPS, average latency (ms) and average queue depth.
  2. Press Calculate.

Fields

IOPS
Completed I/O per second.
Average latency
Milliseconds; esxtop DAVG/GAVG or Windows 'Avg. Disk sec/Transfer' × 1000.
Queue depth
Average outstanding I/O.

Worked example

InputValue
IOPS2000
Average latency5

Expected result: 2,000 IOPS × 5 ms = on average 10 outstanding I/Os.

At queue depth 10 the disk or array delivers 5 ms latency. A system delivering the same IOPS at 1 ms would have a queue of 2.

Second scenario

InputValue
Average latency0.8
Queue depth32

Expected result: 32 concurrent I/Os at 0.8 ms latency → up to 40,000 IOPS.

Method and assumptions

Queue depth = IOPS × latency (s). Steady state is assumed; with three values a deviation above 2 % is reported as an error.

Using the result and troubleshooting

Latency above 20 ms with a high queue means storage is saturated; low queue with high latency points to the path (HBA, network, controller).

Limits and common mistakes

  • Works with averages; bursts and percentile (p99) latencies are not visible.

Frequently asked questions

What is a good latency?

All-flash arrays typically below 1 ms, hybrid and spinning disks 5–20 ms. Above 20 ms users notice.

Related tools and service

Related service: Hardware Consulting

Systems online tools · See all online tools