Score:0

Linux -LMV DM devices IOPS count

cn flag

I know this might sound like a silly question, but here I go. I have CentOS 7 running in a server with a disk using LVM, according to the vendor the disk could achieve 500 iops, the disk is under /dev/sda and I can see the LVM DM devices showing up as dm-0 and dm-1. I'm monitoring the server with telegraf and plotting the metrics on Grafana to see the behavior of multiple metrics when I run my applications, up to this point all is good. The question I have is regarding the iops count, should I count the both the sda and dm-x io request or just sda to measure my average io request usage?

Grafana sample

Score:1
kr flag

The LVM devices starting with DM are an abstraction on top of the disk device.

The disk device is /dev/sda.

Your vendor offered IOPS values for the disk itself.

You probably should measure the performance of /dev/sda if you want to evaluate the claims of the vendor.

Note that the "average" I/O performance is unlikely to closely approach the peak rate unless your disk usage is consistent and saturated, at which point your experience of the performance of the system is unlikely to be positive (it implies that some process is waiting for I/O literally all_the_time.).

Escanoru avatar
cn flag
Thanks for providing your feedback, it sounds like the iops from the LVM DM devices are "artificial", are the DM iops actually presenting a load to the system?
Escanoru avatar
cn flag
Looking at the metrics shared in the following screenshot I see quite the activity for both RIOPS and WIOPS on the both devices: https://imgur.com/a/hYbDUTF
kr flag
Those images are interesting. It looks to me like the LVM setup is providing some level of write reduction (could be coalescing multiple writes into one?) but not much read reduction (you only show one of the two LVM devices in the graph, so hard to be sure).
mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.