rrqm/s & wrqm/s always zero

Latest response

Hello,

On RHEL 6, iostat -dkNx 2 shows that rrqm/s & wrqm/s is always zero. What could be the reason?

Its a 12c RAC system with disks in ASM and the server is pretty busy.

I am suspecting that either iostat is reporting incorrectly, possibly due to a bug OR may be there is some performance configuration (may be IO scheduler bug or driver bug) because of which this is happening.

Does anybody have experienced this?

Thanks

PC
Log in to join the conversation

Responses

SM Guru 3731 points
7 August 2018 11:56 AM

Those parameters i.e. "rrqm/s" & "wrqm/s" would show 'number of read requests merged per second that were queued to the IO scheduler for a device'. So, by default in RHEL6.x the default IO scheduler is CFQ (Completely Fair Queuing) which is ideal for a system which does a lot of small disk read/writes. So you may have to check what elevator is being used. To find out the elevator being used on a disk : # cat /sys/block//queue/scheduler

Christian Labisch's picture Guru 9710 points
7 August 2018 12:27 PM

A little correction Sadashiva ... the disk is missing : cat /sys/block/<disk (e.g. sda)>/queue/scheduler
Hope you don't mind ... :)

Regards,
Christian

PC Community Member 45 points
7 August 2018 12:08 PM

Thanks for your response. Thats exactly the query that why are reads/writes not being merged? I understand that if its small random reads/writes it would merge less but to always show zero makes me doubt.

The elevator in use is indeed CFQ. I have been researching and various KB on RHN (e.g. KB 39188) suggest that Deadline is almost always better suited for databases.

However, when I look at KB 54164, the symptons that should be seen for Deadline to be a fit are not seen.

Thanks

PC Community Member 45 points
7 August 2018 1:02 PM

Yes Sadashiva, I have gone through that KB and several others too. But I was looking for a way if stats from production system can help confirm if the change will really help. In other words, if stats can prove that CFQ is not doing a good job. So I started analyzing the iostat o/p to see if there are any issues. One of the observation was and rrqm/s & wrqm/s is always zero. I am not sure if that is related to Deadline/CFQ. Any other way to verify from the current stats before going for the change?

Thanks

Close

Welcome! Check out the Getting Started with Red Hat page for quick tours and guides for common tasks.