Is there any performance counter available to determine whether the workload is more of sequential read or random read ? we are facing a disk contention in FAS 8300 due to high read request , would like to narrow down the workload using it.
#Performance metrics to capture the IO pattern for specific SVM /Volume
elder iris
spring furnace
can you create a case and get a perfstat?
rotund pendant
pstejska_vsim::*> statistics start -object workload_volume -counter sequential_reads|sequential_writes
Statistics sample deleted, for sample-id: sample_30
Statistics collection is being started for sample-id: sample_30
pstejska_vsim::*> statistics show
Object: workload_volume
Instance: lun1-wid7398
Start-time: 6/28/2023 11:28:29
End-time: 6/28/2023 11:48:47
Elapsed-time: 1218s
Scope: pstejska_vsim_node
Number of Constituents: 1 (complete_aggregation)
Counter Value
-------------------------------- --------------------------------
sequential_writes 99%
pstejska_vsim::*>
elder iris
Thanks, this counter works . How do we intepret this output ,for the most of the volumes , i got sequential writes value as 99 % and some with very minimal sequential read , but if we sum up it does count to more than 100% .
rotund pendant
It may not be ever 100%.
elder iris
sorry , sum up goes more than 100% . if it's 99% sequential write does that mean it's predominantly sequential write than sequential read
rotund pendant
Writes and reads are two different types of requests, so it's possible to have 1% sequential reads and 99% sequential writes or vice versa.
elder iris
I know both are different. I am getting this kind of output for many volumes with the mix of reads and writes ( which doesn't match with 100%)
Object: workload_volume
Instance: aas-wid3043
Start-time: 7/11/2023 12:12:22
End-time: 7/12/2023 17:25:19
Elapsed-time: 105177s
Scope: svbb001
Number of Constituents: 2 (complete_aggregation)
Counter Value
-------------------------------- --------------------------------
sequential_reads 66%
sequential_writes 99%
rotund pendant
It might be different when you look across all volumes vs. specific volumes.
It depends on the quantity of reads and writes in each volume.
elder iris
ok , So how do we conclude the pattern on each volume based on this counter value Counter Value
-------------------------------- --------------------------------
sequential_reads 66%
sequential_writes 99%
rotund pendant
I guess I don't understand the question? That shows you the read/write sequential values for that specific volume.
hollow scaffold
The output should be interpreted to mean that, during the time those statistics were collected, 66% of the total reads were sequential, and 99% of the total writes were sequential
burnt locust
ok , So how do we conclude the pattern on each volume based on this counter valu...
66% of all reads are sequential (i.e. 34% of reads are random). At the same time, 99% of writes are sequential (with 1% of writes being random). I'm not sure what to tell you, because I also don't quite understand the question? The counter works as intended, I'm not sure what you are looking for
elder iris
Thanks all, it's clear now. i misunderstood the counter definition