Error in post-processing tar file about ioprof HOT 10 CLOSED

ioprof leverages blktrace and blkparse under the hood. So, you should be able to capture IO of ramdisk devices or the underlying block devices that you mmap. I suspect that the tool is having issues with the fdisk output on your systems. I have a Ubuntu 18.04 system available if you'd like me to try reproducing something similar to your issue

from ioprof.

Thanks for the help. I'm attaching the mmap code. It does reading and writing. It's really simple in that it reads a text file, changes every other character to a uppercase, which should result in writing to disk. Pick any text file you want, but make it somewhat lengthy to get some measurable file IO :) I used the constitution at first and then used a massive wikipedia file.

The code is in C but I can't upload C code, so I just changed it to a text extension.

Thanks!
memory-mapped-test-write.txt

from ioprof.

Are you pointing ioprof.pl to your text file? The tool only works on block devices (i.e. /dev/sda1), so you would need to figure out what filesystem your file resides on and which disk/partition it is mounted on. So, if your file is /home/laytonjb/NVIDIA/IO_PROFILE/ioprof/sometext.txt, then you'd need to run mount to figure out where /home is mounted. For example, if / (and thus /home) are mounted on /dev/sda1, you would use /dev/sda1 as the device to trace.

/dev/sda1 on / type ext4 (rw,relatime,errors=remount-ro)

from ioprof.

from ioprof.

from ioprof.

Great, thanks for the additional information. It should create a file within the .tar file called fdisk.sdb1 with the output of either 'fdisk -ul /dev/sdb1' OR 'fdisk -l -u=sectors /dev/sdb1', depending on the output of fdisk -v. I will try it out on my Ubuntu 18.04 LTS system today to see if the fdisk output is unexpected.

from ioprof.

I have reproduced the issue and it appears that the output of fdisk has changed a bit on Ubuntu. I've pushed a quick patch to the repo for ioprof.pl. Please pull the latest version and re-run your post-processing job on the same .tar file.

I will let you know that the tool won't have much of a heatmap to show for a 4 second runtime. It works much better with larger datasets like minutes or hours.

from ioprof.

from ioprof.

The quick patch I made checked the $sector_size value to see if it matched "bytes" instead of "512", as expected. That said, it needed the "\n" newline chomped off and to use string, rather than numerical comparisons. I've pushed a patch for this as well, but plan to review the new fdisk output format in greater detail and push a more reliable fix later this month.

from ioprof.

from ioprof.