View Issue Details

IDProjectCategoryView StatusLast Update
0017078CentOS-7kernelpublic2020-03-04 10:57
Reportermiw Assigned To 
Status newResolutionopen 
PlatformGCEOSCentos7OS VersionLinux 3.10.0-106
Product Version7.7-1908 
Summary0017078: High await times for IO operations on single queue devices.
DescriptionWe are observing high await/w_await/r_await values (in our case, up to ~150ms, about 8x higher result when ran on same setup with CentOS 8) reported by iostat for disk IO operations. The issue can be easily reproduced by creating a sample file, decompressing it using gzip and running "iostat -x 1" in parallel. When the benchmark was changed to write from /dev/zero to a file the results are consistent. It suggests gzip is not important here.

Our setup was running inside virtual machine with 16 cores with 16gb memory (to make sure that there are no other constrains) and IO tests were running on single queue (virtual) SCSI disk. This issue is visible with Centos7 but not with CentOS 8.
Steps To Reproduceyum install -yq sysstat
head -c 2G </dev/urandom > 2G.bin
iostat -x 1 > iostat.txt &
sleep 10
time gzip -c -d -f 2G.bin > 2G.bin.gz
sleep 10

# high await/w_await/r_await values should be visible in IO stat output.
cat iostat.txt
TagsNo tags attached.


There are no notes attached to this issue.

Issue History

Date Modified Username Field Change
2020-02-21 18:12 miw New Issue