We are using Vdbench version 5.04.07 to conduct tests on enterprise storage systems. The following configuration is used:
sd=sd1,lun=/dev/dm-3,openflags=o_direct
wd=wd1,sd=sd1,xfersize=32k,rdpct=50,seekpct=100
rd=rd1,wd=wd1,iorate=50000,threads=64,elapsed=86400,warmup=10
The tests include various stress scenarios, including the impact on the hardware components of the storage system. During certain tests, the storage system may temporarily halt I/O processing. This is expected behavior. We use a fixed iorate value to observe how each test scenario affects the host generating the test load. After the I/O operations resume, the storage system returns to normal functioning. However, a spike effect is observed: instead of maintaining a steady IOPS level, Vdbench appears to "catch up" by sending the I/O operations it was unable to issue while the storage system was not accepting them.
So I have two questions.
Is this behavior normal for Vdbench? If so, is there a way to change it so that when I/O resumes, Vdbench simply returns to generating the target IOPS level as specified by the iorate parameter, instead of trying to "catch up"?
