I'm assuming that was a run in the foreground and not using QoS? Our timings sound roughly similar for a Foreground run under 4.2.3.x. 1 hour and ~2 hours for 100million and 300 million each. Also I'm assuming actual file counts, not inode counts! Background is, of course, all over the place with QoS. I've seen between 8-12 hours for just 100 million files, but the NSDs on that FS were middling busy during those periods.
I'd love to know if IBM has any "best practice" guidance for running mmcheckquota. Ed ________________________________ From: [email protected] <[email protected]> on behalf of Oesterlin, Robert <[email protected]> Sent: Monday, August 19, 2019 9:54 AM To: gpfsug main discussion list <[email protected]> Subject: Re: [gpfsug-discuss] Running mmcheckquota on a file system with 1.3B files Thanks, - I kicked it off and it finished in about 12 hours, so much quicker than I expected. Bob Oesterlin Sr Principal Storage Engineer, Nuance From: <[email protected]> on behalf of IBM Spectrum Scale <[email protected]> Reply-To: gpfsug main discussion list <[email protected]> Date: Monday, August 19, 2019 at 8:24 AM To: gpfsug main discussion list <[email protected]> Subject: [EXTERNAL] Re: [gpfsug-discuss] Running mmcheckquota on a file system with 1.3B files Bob, like most questions of this time I think the answer depends on a number of variables. Generally we do not recommend running the mmcheckquota command during the peak usage of your Spectrum Scale system. As I think you know the command will increase the IO to the NSDs that hold metadata and the number of NSDs that hold metadata will contribute to the time it takes for the command to complete, i.e. more metadata NSDs should improve the overall execution time. Regards, The Spectrum Scale (GPFS) team
_______________________________________________ gpfsug-discuss mailing list gpfsug-discuss at spectrumscale.org http://gpfsug.org/mailman/listinfo/gpfsug-discuss
