Standard test parameters for NAS workloads?

Standard test parameters for NAS workloads?

[rickyp999]

Hi,

I'm curious whether there are some standard or at least common test
parameters for benchmarking NAS workloads.

If NAS is too specific, even a general one for HDD/SSD would be
greatly appreciated.  My motivation is to use a common set of
benchmarks to compare how options like RAID levels, network
speed/protocol (through a NFS/SMB mount), and SSD cache affect
performance.

Thank you,

Re: Standard test parameters for NAS workloads?

[Davor Cubranic]

These are the fio settings used by StorageReview.com in their benchmarks: https://www.storagereview.com/fio-flexible-i-o-tester-synthetic-benchmark

Ars Technica’s series on RAID vs. ZFS (https://arstechnica.com/gadgets/2020/05/zfs-versus-raid-eight-ironwolf-disks-two-filesystems-one-winner) used this set of tests: https://github.com/jimsalterjrs/fio-test-scaffolding



> On Dec 21, 2020, at 2:49 PM, rickyp999@gmail.com wrote:
> 
> Hi,
> 
> I'm curious whether there are some standard or at least common test
> parameters for benchmarking NAS workloads.
> 
> If NAS is too specific, even a general one for HDD/SSD would be
> greatly appreciated.  My motivation is to use a common set of
> benchmarks to compare how options like RAID levels, network
> speed/protocol (through a NFS/SMB mount), and SSD cache affect
> performance.
> 
> Thank you,

Re: Standard test parameters for NAS workloads?

[rickyp999]

These are great articles. Thank you so much and I'll incorporate some
of these tests into my configurations.

One more question if that's ok, the articles did not really mention
how they handled networking. Looking at the commands and the test
scaffolding, it seems like these tests ran directly against the disks
locally. I'm curious if there's anything one should do differently
when testing against them on networked filesystems such as SMB or NFS,
like whether one should use different settings compared to testing
against local disks with regards to. For example, I don't know how
buffered I/O works on networked filesystems and whether to turn it on
or off in the tests.

Thank you,

On Mon, Dec 21, 2020 at 5:21 PM Davor Cubranic <davor@phas.ubc.ca> wrote:
>
> These are the fio settings used by StorageReview.com in their benchmarks: https://www.storagereview.com/fio-flexible-i-o-tester-synthetic-benchmark
>
> Ars Technica’s series on RAID vs. ZFS (https://arstechnica.com/gadgets/2020/05/zfs-versus-raid-eight-ironwolf-disks-two-filesystems-one-winner) used this set of tests: https://github.com/jimsalterjrs/fio-test-scaffolding
>
>
>
> On Dec 21, 2020, at 2:49 PM, rickyp999@gmail.com wrote:
>
> Hi,
>
> I'm curious whether there are some standard or at least common test
> parameters for benchmarking NAS workloads.
>
> If NAS is too specific, even a general one for HDD/SSD would be
> greatly appreciated.  My motivation is to use a common set of
> benchmarks to compare how options like RAID levels, network
> speed/protocol (through a NFS/SMB mount), and SSD cache affect
> performance.
>
> Thank you,
>
>

Re: Standard test parameters for NAS workloads?

[Davor Cubranic]

I don’t know, and would myself be interested to hear what you learned!

Davor

> On Dec 22, 2020, at 7:28 PM, rickyp999@gmail.com wrote:
> 
> One more question if that's ok, the articles did not really mention
> how they handled networking. Looking at the commands and the test
> scaffolding, it seems like these tests ran directly against the disks
> locally. I'm curious if there's anything one should do differently
> when testing against them on networked filesystems such as SMB or NFS,
> like whether one should use different settings compared to testing
> against local disks with regards to. For example, I don't know how
> buffered I/O works on networked filesystems and whether to turn it on
> or off in the tests.