Story concepts: Throughput, latency and IOPS

These are three important parameters in storage systems. In order to understand these three concepts, it is possible to map with shipping operations from point A to B.

The number of current trips over a period of time is IOPS

The number of goods delivered within a period of time is the throughput

latency is the average delay in all trips in a given time period

This time is supposed to be a day away.

These three parameters, especially the two IOPS parameters and latency reflect the service quality but not always go together in a good indicator, the remaining indicators are also good according to:

There may be many shipments a day but there are fast shipments, slow shipments, and high IOPS, but the average latency is also high.

There may be a few shipments a day but each load carries a full load, a high throughput though low IOPS because Throughput = IOPS * IO Average size (high IO average size means a high throughput)

It is possible that the average latency is low, but the number of transfers is not so high because of the few orders (application less requests to storage)

But not so that these parameters have no effect on each other:

When the IOPS is too high, touching the system's physical limit will cause high latency

non-processing high latency will reduce throughput because data is not actually transferred to the right place but is blocked (busy is also high)

You can view these three parameters via the atop tool:

Screen Shot 2016-08-04 at 10.01.00 AM

where the line DSK sda, busy 1%, read 0, write 339 so IOPS = 339, MBread / s = 0, MBWrite / s = 0.27, throughput write is 0.27MB / s, and avio is the average latency of 0.40ms.

I am learning about storage. Understanding the concept is not exactly what you comment to edit.

Written by Manhdung

ITZone via Kipalog

Share the news now