Does diskspd create "real" sparse files? / Behaviour when using SMB

[mt1102]

Hello,

(i hope its ok to ask a question like this here)

Can you help me explain the following behaviour of diskspd:

When i use this to work with a local file:

diskspd.exe -c10G -w50 -b32K -F8 -r32K -o32 -W60 -d60 -Sh -L c:\Testfile

the results are as expected.

When i use this to work with a remote file:

diskspd.exe -c10G -w50 -b32K -F8 -r32K -o32 -W60 -d60 -Sh -L \servername\c$\Testfile

(servername is a VM hosted on the same ESX in a vSphere environment so i guess i can rule out network issues etc.)

the following happens:

• The results are very bad, much worse than expected.

• The 10G file is created immediately on the remote server. So i guess it is a sparse file? However, fsutil says it is not marked as sparse file (command is use: fsutil sparse queryflag on the SMB server locally).

• I can see in the taskmanager of the client that almost no network traffic is happening, even beyound the 60s waiting delay. This is different when i use a much smaller file - for example 5M - then i can see fast network traffic immediately.

Do you have an explanation for this behaviour? It seems to me that the way the file is created causes measurements over SMB access to show odd results.

[dl2n]

DISKSPD does not create sparse files.

What I'd expect is this warning, though (and I can go into details, but first). You don't get this? Is the VM Windows?

WARNING: Could not set valid file size (error code: 1314); trying a slower method of filling the file (this does not affect performance, just makes the test preparation longer)

If you don't, I'd be really curious and expect that if you pop Task Manager on the server VM, you'll see a lot of write activity occurring.

[mt1102]

Hello.

No, i didnt get this warning and yes, both VM's are Windows 2022 Servers.

However, i get this warning when i use one of the Windows 2022 Server als "Client" (that runs the diskspd.exe) and a QNAP NAS storage with a SMB share. The results then are good, way more better than using Windows 2022 as Server.

I dont remeber if i noticed a lot of write activity as you describe it (its been a while since i last used it), i can try to check if you want. Is there any specific parameter in performance monitor to look for?

But i realized that the -c10G parameter seems to have a great influence. When i use it as described above the results are very bad over SMB.

However, i tried the following:

1. Use diskspd as mentioned above:

diskspd.exe -c10G -w50 -b32K -F8 -r32K -o32 -W60 -d60 -Sh -L \servername\c$\Testfile

(the file Testfile is created within a fraction of a second on the Server)

2. Then i switch to Server "servername" and execute diskspd with the following parameters against the created file "Testfile":

diskspd -w100 -Zr -d180 c:\Testfile

3. Then i switch back to the client and execute diskspd again without the -c parameter, now the results are as expected (worse than writing to a local file on disk but still ok)

So it seems to me the "empty" file created with -c10G parameter somehow leads to odd results. The storage that is used in our vSphere environment uses both compression and deduplication, maybe this has to do with it.