Skip to contentSkip to conversation

fio terse output Write/Read Bandwidth (mean) = zero depending on ...blocksize, filesize?

DISCUSSION mmoole 2 participants 2 messages Apr 17, 2023

lore
Share & tools
lore

mmoole

author
#0 lore raw

Dear fio Team,

I am running fio in a loop with varying values for some parameters:
FIO_RESULT=$(fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --bs=$BLOCK_SIZE --iodepth=64 --readwrite=randrw --rwmixread=75 --sync=sync --numjobs=5 --group_reporting --size="$TEST_FILE_SIZE" --filename="/$ZPOOL_NAME/$TEST_FILE" --output-format=terse)

For each run a zfs pool is created and afterwards discarded.

Now the output I analyse in Excel looks strange at some points:
* some combination of those values lead to output of a lot of zeros in relevant columns, especially Write Bandwidth (mean) and Read Bandwidth (mean) and their accompanying columns.
* it seems that small test file size and large block size lead to that, but not exclusively?
* on the other hand Write Bandwidth (KB/s) column gives huge values compared to other runs, how does that make sense?

See picture of collapsed table attached. (https://imgur.com/a/V5FsmmU )

Would it help to provide the full table, or is there a logical explanation for this, like something regarding block size, multipliers, filesize, r/w percentage or so on?

greetings
Sven

#1 lore raw
Quoted 16 lines · mmoole

Dear fio Team,

I am running fio in a loop with varying values for some parameters:
FIO_RESULT=$(fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --bs=$BLOCK_SIZE --iodepth=64 --readwrite=randrw --rwmixread=75 --sync=sync --numjobs=5 --group_reporting --size="$TEST_FILE_SIZE" --filename="/$ZPOOL_NAME/$TEST_FILE" --output-format=terse)

For each run a zfs pool is created and afterwards discarded.

Now the output I analyse in Excel looks strange at some points:
* some combination of those values lead to output of a lot of zeros in relevant columns, especially Write Bandwidth (mean) and Read Bandwidth (mean) and their accompanying columns.
* it seems that small test file size and large block size lead to that, but not exclusively?
* on the other hand Write Bandwidth (KB/s) column gives huge values compared to other runs, how does that make sense?

See picture of collapsed table attached. (https://imgur.com/a/V5FsmmU )

Quoted 8 lines

Would it help to provide the full table, or is there a logical explanation for this, like something regarding block size, multipliers, filesize, r/w percentage or so on?

greetings
Sven

Fio will report two types of bandwidth figures:

(1) overall bandwidth: (total bytes) / (total time)
(2) average of bandwidth values from periodic samples taken during the job run

You are observing a situation where (1) is non-zero but (2) is zero.

All the evidence you have presented suggests that your jobs are finishing too quickly and fio has not collected any bw sample data and reports zeros for the sample-based statistics.

For more details see:

https://fio.readthedocs.io/en/latest/fio_doc.html#interpreting-the-output

Vincent