Operation Failed! when deleting a Virtual Disk in EVA4100

[Uwe Zessin]

I'm not sure I understand your question.

While VMFS does support sub-block allocation (can store multiple smaller files in a single VMFS block), the VMFS blocksize has nothing to do with the guest OS blocksize within the VMDK. And the VMkernel does not do de-blocking with 8MB read/write I/Os - it forwards the guest's I/O request which is often way smaller. You can ask EVAperf to show a distribution of the I/O sizes.


I don't think that building a VMFS datastore with multiple extents is a good idea unless one is forced by the storage array, because there is still one I/O wait queue (throught the first partition).
And with ESX4 and a competent array like the EVA you can even grow an existing VMFS partition and the embedded filesystem.

If you require more I/Os it might be necessary to increase some queue depths or create multiple VMDKs (and then multiple filesystems) for the fileserver. The downside is that free space is now scattered, but it could also help to parallelize some backups.

Yes, I'm german.

[twin]

That was completely new to me, thank you.

Since we are offtopic anyway some last words in german because my english is not sufficient:

Das heisst also, es ist völlig egal welche Block Size mein VMFS3 Volume hat, weil ohnehin nur das VMDK-File abgelegt wird und das Filesystem des Guests innerhalb des VMDK liegt. Auf einem Fileserver liegen üblicherweise sehr viele deutlich kleinere Files als 8MB, würden aber nicht jedesmal einen 8MB-Block verschwenden, weil die Filesysteme völlig unabhängig voneinander sind ?
Ich kann also demnach problemlos ein 1,8TB grosses VMFS-Volume anlegen mit einer Blockgrösse von 8MB und dieses dann auch als ein grosses Volume im Guest anlegen ohne Verlust bei vielen kleinen Dateien.
Das wäre klasse, weil ich gern unter Windows einen dynamischen Datenträger anlegen möchte mit der Anfangsgrösse von 1,8TB und dieses später ggfls. mit weiteren Volumes erweitern möchte.

[Uwe Zessin]

Perfect. Remember that VMFS partitions and VMDK are limited to 2 TeraBytes. If you need a larger NTFS filesystem you do need to map multiple VMDKs to a VM and then combine them with the guest OS' volume manager.

While you can create a VMFS datatore larger than 2TB by using extents I would not do so - due to a single I/O queue to the entire filesystem.

[twin]

Thanks a lot Uwe !