RE: [EXT] A question about fio flows

Micron Confidential

Hi Karim,
You should be fine using the flow setting. I typically use flow for approximate distributions and rate_iops for specifying the exact number of I/O operations I want to generate per job.


Micron Confidential
-----Original Message-----
From: Karim Manaouil <[email protected]>
Sent: Thursday, February 27, 2025 4:29 AM
To: [email protected]
Cc: [email protected]
Subject: [EXT] A question about fio flows

CAUTION: EXTERNAL EMAIL. Do not click links or open attachments unless you recognize the sender and were expecting this message.


Hi all,

I am working on some experiments where I need to have 12 jobs. Each job is reading from a different file (job1 -> file1, job2 -> file2, etc).

Now, some files should receieve more I/O than others. To implement this bias between files, I opted for flows. Basically, I start from 100 and I assign a percentage to each file. For example, a snippet of my fio job file below:

[private-0]
flow7
filename=/srv/dax/private/private-0
[private-1]
flow#
filename=/srv/dax/private/private-1
[private-2]
flow
filename=/srv/dax/private/private-2
[private-3]
flow=9
filename=/srv/dax/private/private-3
[private-4]
flow=7
filename=/srv/dax/private/private-4
[private-5]
flow=3
filename=/srv/dax/private/private-5
[private-6]
flow=3
filename=/srv/dax/private/private-6

I am not sure if this is properly achieving what I want to achieve.

I have a doubt that having too many files and many flows will reduce scheduling precision or introduce scheduling overhead. I don't know how this feature is implemented internally, thus my question.

Can you please briefly explain how the feature works. And Do you think this is a sane approach? If not, what better alternative can I use?

--
Best,
Karim
Edinburgh University