How to run the SLOG benchmark under Scale?

[rdfreak]

I want to run the SLOG benchmark from this thread, but I've encountered a couple of issues:

1. My main NAS is TrueNAS Scale, and it doesn't seem to include the diskinfo tool
2. I temporarily installed TrueNAS Core, and though diskinfo is present, it fails with an error:

Code:

Synchronous random writes:
0.5 kbytes: diskinfo: Flush error: Inappropriate ioctl for device​

​

​

My request for help is to address either of the two issues, so that I can post my benchmark results and contribute with data for a drive that so far hasn't had its performance as SLOG device tested:
What alternative tool can I use on TrueNAS Scale or how do I install diskinfo? or How can I fix that ioctl error?

So far, I tried creating a dataset with sync=always and dd-ing from /dev/random with various block sizes. Despite the ioctl error earlier, the drive can be written to, and it doesn't seem too bad, but I don't know whether that's good enough. And I don't see any latency numbers.

 

[jgreco]

TrueNAS Scale, and it doesn't seem to include the diskinfo tool

Linux likes to reinvent the wheel and do their own thing.

Despite the ioctl error earlier, the drive can be written to, and it doesn't seem too bad, but I don't know whether that's good enough. And I don't see any latency numbers.

Without some idea as to what device it is that you have, it's difficult to guess as to what happened.

 

[morganL]

does the device support 512B writes or only 4kB?

 

[jgreco]

does the device support 512B writes or only 4kB?

Yeah, the inappropriate ioctl was kinda making me think along those lines, but I wasn't exactly interested in delving into the diskinfo code without some idea of what to look for.

 

[rdfreak]

Thank you for responding, @morganL and @jgreco.

The drive is a Samsung PM9A3 NVMe SSD, U.2 version. Supports 512-byte writes, though I doubt that's the page size.

The unusual thing is that I created 2 NVMe non-shared namespaces, and I wanted to test only on one them; the other one accommodates TrueNAS Core temporarily. This doesn't stop the drive and second namespace to be useful - I managed to create a fully functional pool with the second namespace, and I ran some tests with dd (which I now know doesn't get around file system overheads), and fio. Both managed to write using block sizes varying from 512 bytes to 32 KiB. However, I ran the latter tool using parameters I thought were relevant (using sync or sync anywhere where the option presented itself) , so I don't know for sure whether I ran the benchmark correctly for Slog purposes. Nor whether diskinfo does something the other commands don't, which might explain the error.