Re: Aggregated results too high when using multiple servers via "client" command

Unfortunately not. But I did some more research and my best guess at the moment is that FIO is correct and I am only testing the caching engine on the target.
I did some restore tests as well which did also over perform but I did see barely disk access on the the target site. 
My target machine is a Linux with a ZFS File System, connected via SMB. The system comes with 128GB RAM. 
I tried increasing the filesize to over 128GB which reduces the summarized performance on write but not on read.
I tried mounting the volume with gifs option cache=none without effect.
Funny enough when not using multi-client but 6 parallel jobs I get the same too-high values as initially with multi-client configuration.

So I guess it is rather a general setup problem and less a bug. Nevertheless I would be happy to get a good advice...



> Am 05.05.2025 um 23:33 schrieb Sitsofe Wheeler <[email protected]>:
> 
> On Mon, 5 May 2025 at 11:56, René Weber <[email protected]> wrote:
>> 
>> You are absolutely right. Lots of details missing indeed…
>> 
>> fio-3.36 on Ubuntu 24.04.2 LTS
>> 
>> Command:
>> fio --client=configs/host.list fio-configs/test.job --output=test.output --output-format=json
>> 
>> cat test.job:
>> [global]
>> filename=/mnt/benchmark/testfile.dat
>> group_reporting
>> write_bw_log=aggregated
>> 
>> [Full Write]
>> size=25G
>> bs=512K
>> rw=write
>> ioengine=libaio
>> direct=1
>> time_based
>> runtime=60s
>> 
>> cat host.list
>> 10.100.12.51
>> 10.100.12.52
>> 10.100.12.53
>> 10.100.12.54
>> 
>> My target is a NAS device mounted via CIFS/SMB under /mnt/benchmark/
>> I played around with write_bw_log=aggregated and group_reporting, but in the end the summarized Bandwidth  is always higher than my target device overall performance.
> 
> Is there any difference in performance numbers if you add end_fsync=1
> to the job file?
> 
> -- 
> Sitsofe