StarWind Software

Hi,

Just installed v8.0.6441
How do i re-mount these LSFS
Thanks

Just another minor note for RC, I created a drive with an L2 cache, and forgot to change the location that the L2 cache was stored on. It happily accepted my primary storage as the location. It probably wouldn't hurt to add a warning if you're trying to put L2 cache on the same location as your primary storage, since I would assume if primary storage was SSD, you wouldn't be using L2 cache as well on it.

fbifido wrote:Hi,

Just installed v8.0.6441

iscsi_help8.png

How do i re-mount these LSFS

iscsi_help9.png

Thanks

Since RC LSFS format changed, so update from the beta versions is not supported. Please just re-create LSFS device.
Build and release notes are available to download on this page http://www.starwindsoftware.com/starwind-v8-beta

jcrowe wrote:Just another minor note for RC, I created a drive with an L2 cache, and forgot to change the location that the L2 cache was stored on. It happily accepted my primary storage as the location. It probably wouldn't hurt to add a warning if you're trying to put L2 cache on the same location as your primary storage, since I would assume if primary storage was SSD, you wouldn't be using L2 cache as well on it.

this will be improved soon by setting default pool for L2 cache with automatic SSD detection.

Since RC LSFS format changed, so update from the beta versions is not supported. Please just re-create LSFS device.
Build and release notes are available to download on this page http://www.starwindsoftware.com/starwind-v8-beta

Noted. Thanks.

My observations testing v8 RC

- Deduplication is CPU intensive. I observed 50% cpu-load on my single i3 Xeon 2.5 GHz (2-core 2.5GHz + HT) during 50MB/s writes
Is it not possible to adjust dedupe block size any more? Found only checkbox for auto-defrag and dedupe on/off in LSFS options.

- UNMAP works! (vSphere 5.5). Cloned/migrated 10 VM's to LSFS volumes (with and without dedupe), and when deleting or migrate away VMs I was able to unmap using this command on the vSphere host:
esxcli storage vmfs unmap -l "datastore name"
After all data deleted from datastore + unmap = only single 1GB spspx on SW-box (VMFS metadata)

VAAI mostly works.

- Clone from IMAGE to LSFS:
Writes observed in ESXTOP (MBWRTN/s), but almost not reads . VAAI-stats show MBC_RD/s and MBC_WR/s
Network on SW-box confirm (receive traffic, but almost no transmit)

- Clone from LSFS to IMAGE:
All good! VAAI-stats show MBC_RD/s and MBC_WR/s, but almost no network traffic on host and SW-box

- Clone IMAGE to IMAGE:
All good! Same as above

VAAI works both when cloning to same datatore and to other datastores on same SW-box

We keep seamless upgrades from release-to-release but it's not always possible to have easy migration from beta-to-beta. Sorry for this!

fbifido wrote:

Since RC LSFS format changed, so update from the beta versions is not supported. Please just re-create LSFS device.
Build and release notes are available to download on this page http://www.starwindsoftware.com/starwind-v8-beta

Noted. Thanks.

Do you have all cores used? I mean is it real 50% CPU utilization or just one core 50% CPU utilization?

No we'll use 4KB blocks inside and outside. Upcoming upgrade would use 64KB blocks + compression and external 4KB blocks.

For true thin-provisionining you'll have to "trim" the LSFS so it would re-pack the content and squeeze unused data and release space to NTFS it uses to host container files on.

What does it mean "almost"? Issues with writes you've referenced? What other VAAI-enabled implementation (diff. vendor) has no traffic so we could compare.

Thank you!

lohelle wrote:My observations testing v8 RC

- Deduplication is CPU intensive. I observed 50% cpu-load on my single i3 Xeon 2.5 GHz (2-core 2.5GHz + HT) during 50MB/s writes
Is it not possible to adjust dedupe block size any more? Found only checkbox for auto-defrag and dedupe on/off in LSFS options.

- UNMAP works! (vSphere 5.5). Cloned/migrated 10 VM's to LSFS volumes (with and without dedupe), and when deleting or migrate away VMs I was able to unmap using this command on the vSphere host:
esxcli storage vmfs unmap -l "datastore name"
After all data deleted from datastore + unmap = only single 1GB spspx on SW-box (VMFS metadata)

VAAI mostly works.

- Clone from IMAGE to LSFS:
Writes observed in ESXTOP (MBWRTN/s), but almost not reads . VAAI-stats show MBC_RD/s and MBC_WR/s
Network on SW-box confirm (receive traffic, but almost no transmit)

- Clone from LSFS to IMAGE:
All good! VAAI-stats show MBC_RD/s and MBC_WR/s, but almost no network traffic on host and SW-box

- Clone IMAGE to IMAGE:
All good! Same as above

VAAI works both when cloning to same datatore and to other datastores on same SW-box

A little followup

I created a 100GB LSFS LUN and created a VMFS datastore. I created a 50GB thick provisioned (lazy zeroed) on a VM (on that datastore)
This is not using any space on the SW-box, but it uses 50GB on the vSphere datastore (as expected)

I copied 40GB of data to the VM disk. I see many 1GB spspx files on the SW-box (as expected)

Now I deleted all data and ran sdelete. I saw almost no data transfer (nic) during sdelete (as it is only zero data), and all spspx-files went away, as I hoped it would do.

So it seems that things work as I hoped/it should.

There are some performance related issues with file pre-allocation but we're working on it. Also we'll try to do "native UNMAP" so there would be no need for sdelete. Thank you!

lohelle wrote:A little followup

I created a 100GB LSFS LUN and created a VMFS datastore. I created a 50GB thick provisioned (lazy zeroed) on a VM (on that datastore)
This is not using any space on the SW-box, but it uses 50GB on the vSphere datastore (as expected)

I copied 40GB of data to the VM disk. I see many 1GB spspx files on the SW-box (as expected)

Now I deleted all data and ran sdelete. I saw almost no data transfer (nic) during sdelete (as it is only zero data), and all spspx-files went away, as I hoped it would do.

So it seems that things work as I hoped/it should.

Good

"native unmap" would problably only work on LUNs directly attached (ISCSI in OS) to guest VMs running Server 2012 > on vSphere, and not on disks on VMFS.

Hyper-V 2012 will allow guests to "send" unmap all the way I think.

Or am I wrong?

Combination of both actually.

lohelle wrote:Good

"native unmap" would problably only work on LUNs directly attached (ISCSI in OS) to guest VMs running Server 2012 > on vSphere, and not on disks on VMFS.

Hyper-V 2012 will allow guests to "send" unmap all the way I think.

Or am I wrong?

anton (staff) wrote:Do you have all cores used? I mean is it real 50% CPU utilization or just one core 50% CPU utilization?

No we'll use 4KB blocks inside and outside. Upcoming upgrade would use 64KB blocks + compression and external 4KB blocks.

For true thin-provisionining you'll have to "trim" the LSFS so it would re-pack the content and squeeze unused data and release space to NTFS it uses to host container files on.

What does it mean "almost"? Issues with writes you've referenced? What other VAAI-enabled implementation (diff. vendor) has no traffic so we could compare.

Thank you!

lohelle wrote:My observations testing v8 RC

I saw 50% total CPU utilization.
"Almost" as in 1% compared to VAAI transfers in the backend. To replicate, clone or svmotion a VM to a LSFS-volume in vSphere 5.5.
ESXTOP
U for storage stats
f-key to select counters, and O (if I remember right) to include VAAI stats. You could also remove som latency-stats for better readability

Then cloning etc using VAAI, there will still be SOME traffic, but this is probably only "control traffic"

That's too high. We'll check this out. Thank you!

Yes, there would be some metadata updates.

Bohdan (staff) wrote: Since RC LSFS format changed, so update from the beta versions is not supported. Please just re-create LSFS device.
Build and release notes are available to download on this page http://www.starwindsoftware.com/starwind-v8-beta

Is there any way for me to access the Beta-3 installer again? I have updated to RC and cannot mount my LSFS device and would like some Data from it, then upgrade it again.

Thanks,
Brad.