Regarding Average Total latency disk I/O

Question

Dear Experts,

I just had a question around disk I/O measurement or latency. Based on the information from DMV's collected, the average total latency appears to be quite high. I came across a link regarding some benchmark measurements.

It says when the average total latency is between 100 ms -500 ms it is considered very bad. On my server it is in that range and I am not sure if I can refer this to my client. Is there any Microsoft reference material regarding this which I can refer?

Thanks,

Answer 1

Hello @mo boy

Welcome to Microsoft Q&A Platform,

The docs below should be helpful i think:

https://learn.microsoft.com/en-us/troubleshoot/sql/performance/troubleshoot-sql-io-performance#define-slow-io-performance
https://techcommunity.microsoft.com/t5/sql-server-support-blog/slow-i-o-sql-server-and-disk-i-o-performance/ba-p/333983
https://learn.microsoft.com/en-us/troubleshoot/sql/admin/io-subsystem-requirements-tempdb

Please have a look

----------

Please don’t forget to "Accept the answer" and “up-vote” wherever the information provided helps you, this can be beneficial to other community members.

Answer 2

Not sure why you would need a Microsoft reference. A standard plain hard disk has a rotation time of 15 ms. Thus, any latency above 15 ms is "fishy". And you are 10 times above that.

And add to that, few SQL Server machines in production these days have such disks. For SSDs, the access time is below 1 ms.

Answer 3

Hi @mo boy ,

Quote from MS document Troubleshoot slow SQL Server performance caused by I/O issues;

>In the world of troubleshooting SQL Server, Microsoft CSS has observed cases where an I/O request took over one second and as high as 15 seconds per transfer-such I/O systems need optimization. Conversely, Microsoft CSS has seen systems where the throughput is below one millisecond/transfer. With today's SSD/NVMe technology, advertised throughput rates range in tens of microseconds per transfer. Therefore, the 10-15 millisecond/transfer figure is a very approximate threshold we selected based on collective experience between Windows and SQL Server engineers over the years. Usually, when numbers go beyond this approximate threshold, SQL Server users start seeing latency in their workloads and report them. Ultimately, the expected throughput of an I/O subsystem is defined by the manufacturer, model, configuration, workload, and potentially multiple other factors.

---

If the answer is helpful, please click "Accept Answer" and kindly upvote it. If you have extra questions about this answer, please click "Comment".