IOPS mismatch with [Max|Min|Mean](OPS)

[santos-lucas]

Hello,

I cant fully understand the big difference between the IOPS value for each iteration and the [Max|Min|Mean](OPS) at the result summary, sorry if it is just my stupidity and not an issue , but looking at the following execution for instance:

IOR-4.0.0: MPI Coordinated Test of Parallel I/O
Began               : Thu Jan 29 20:37:17 2026
Command line        : ior -a posix -F -e -g -b 14g -t 7k -C -i 5
Machine             : Linux c01
TestID              : 0
StartTime           : Thu Jan 29 20:37:17 2026
Path                : testFile.00000000
FS                  : 1039.6 TiB   Used FS: 0.1%   Inodes: 3932.2 Mi   Used Inodes: 0.0%

Options: 
api                 : posix
apiVersion          : 
test filename       : testFile
access              : file-per-process
type                : independent
segments            : 1
ordering in a file  : sequential
ordering inter file : constant task offset
task offset         : 1
nodes               : 64
tasks               : 256
clients per node    : 4
repetitions         : 5
xfersize            : 7168 bytes
blocksize           : 14 GiB
aggregate filesize  : 3.50 TiB

Results: 

access    bw(MiB/s)  IOPS       Latency(s)  block(KiB) xfer(KiB)  open(s)    wr/rd(s)   close(s)   total(s)   iter
------    ---------  ----       ----------  ---------- ---------  --------   --------   --------   --------   ----
write     104035     16924917   0.000006    14680064   7.00       3.56       31.72      3.55       35.28      0   
read      116120     19139176   0.000012    14680064   7.00       3.55       28.05      3.55       31.61      0   
write     102854     15346962   0.000006    14680064   7.00       3.55       34.98      3.55       35.68      1   
read      115588     19039327   0.000013    14680064   7.00       3.55       28.20      3.55       31.75      1   
write     102545     16654813   0.000011    14680064   7.00       3.55       32.24      3.55       35.79      2   
read      117412     19377959   0.000013    14680064   7.00       3.55       27.71      3.55       31.26      2   
write     102603     16665398   0.000012    14680064   7.00       3.55       32.21      3.55       35.77      3   
read      114439     18826171   0.000013    14680064   7.00       3.55       28.52      3.55       32.07      3   
write     107146     17488330   0.000011    14680064   7.00       3.55       30.70      3.55       34.25      4   
read      115905     17901555   0.000010    14680064   7.00       3.55       29.99      3.55       31.66      4   

Summary of all tests:
Operation   Max(MiB)   Min(MiB)  Mean(MiB)     StdDev   Max(OPs)   Min(OPs)  Mean(OPs)     StdDev    Mean(s) Stonewall(s) Stonewall(MiB) Test# #Tasks tPN reps fPP reord reordoff reordrand seed segcnt   blksiz    xsize aggs(MiB)   API RefNum
write      107145.67  102544.72  103836.34    1740.46 15673880.61 15000827.74 15189773.68  254604.69   35.35397         NA            NA     0    256   4    5   1     1        1         0    0      1 15032385536     7168 3670016.0 posix      0
read       117411.81  114439.44  115893.00     955.44 17175670.14 16740855.35 16953489.70  139767.29   31.66943         NA            NA     0    256   4    5   1     1        1         0    0      1 15032385536     7168 3670016.0 posix      0
Finished            : Thu Jan 29 20:42:30 2026

For the summary, I can see that the OPs values makes sense, because they are basically: BW / Transfer size and makes total sense with the code for what I can understand:

static struct results *bw_values(const int reps, IOR_results_t *measured,
                                 const double *vals, const int access)
{
        return bw_ops_values(reps, measured, 1, vals, access);
}

static struct results *ops_values(const int reps, IOR_results_t *measured,
                                  IOR_offset_t transfer_size,
                                  const double *vals, const int access)
{
        return bw_ops_values(reps, measured, transfer_size, vals, access);
}

But, when we are talking about IOPS for each iteration, taking for instance the first write iteration on this test:

access    bw(MiB/s)  IOPS       Latency(s)  block(KiB) xfer(KiB)  open(s)    wr/rd(s)   close(s)   total(s)   iter
------    ---------  ----       ----------  ---------- ---------  --------   --------   --------   --------   ----
write     104035     16924917   0.000006    14680064   7.00       3.56       31.72      3.55       35.28      0

104035 * 1024 = 106531840 (KiB/s for bw)

106531840 (KiB/s for bw) / 7 (xfer KiB) = 15218834,28

What differs from the IOPS value in the result: 16924917

And when I look at the code:

        bw = (double)point->aggFileSizeForBW / totalTime;

        /* For IOPS in this iteration, we divide the total amount of IOs from
         * all ranks over the entire access time (first start -> last end). */
        iops = (point->aggFileSizeForBW / params->transferSize) / accessTime;

I cant really see how those values should be so different, because it seems it is calculated almost in the same way from the summary.