Some understanding of disk IOPS.

It's been 9 months since I said I would write a blog post at 51. Not a thing come out, ashamed!

First write, language organization is not good, forgive me.

In recent years, there have been several problems with SQL Server disk I/O in my work, and finally it comes down to IOPS. I would like to summarize my understanding of IOPS, I hope to help you.

1. The ultimate bottleneck for IOPS is the bottleneck of a single hard disk. Before SSDs, the maximum iops for a single drive is below 200, whether it's a sas,fc,sata drive. SSDs have a single hard drive with ioPS of more than 2000. Therefore, the use of SSD drives in storage can greatly provide the IOPS capacity of this storage subsystem.

2. Write penalty for the raid group (write penalty) issue. RAID 10 is 2,raid 5 is 4,raid 6 is 6. That is, if the write I/O for a server is 400, the total write I/O to the RAID group (assuming all 4 drives) is: Raid 10=800 (400*2), RAID 5=1600 (400*4), and RAID 6 is 2400 (400*6). Assuming that the 15K SAS hard drive (single disk iops=200) is used, there is only a continuous write, so the ioPS per drive is raid10= 200,raid 5=400,raid 6=600. As you can see, the same disk I/O load, the different array types, the hard disk ultimately bear the same IOPS. In the example above, RAID 5/raid 6 may have long Disk queue waits (slow disk response). Therefore, in large and medium-sized systems, try to avoid using RAID 5/raid 6. Otherwise, if you are using high-end storage with dozens of GB of cache, I/O performance issues may also occur.

   
   

It's written here today, and there's time later. How to use the Windows Performance Analyzer to analyze disk I/O monitoring data.

This article is from the "Forest" blog, please be sure to keep this source http://152957.blog.51cto.com/142957/1696396

Some understanding of disk IOPS.