LSFS - Snapshot and async replication - all flash

Software-based VM-centric and flash-friendly VM storage + free version

Moderators: anton (staff), art (staff), Max (staff), Anatoly (staff)

Post Reply
KPS
Posts: 51
Joined: Thu Apr 13, 2017 7:37 am

Wed Dec 19, 2018 8:31 am

Hi!

I did not use LSFS (only raw-devices) in the past, but it seems to be interesting and I need the ability to use snapshots.

Can you tell me something about the limitations?
- Can there be "nearly" unlimited snapshots?
- Is there any possibility to see the space, that is consumed by every snapshot?
- Do you have retention-policies like: Keep 1 hourly snapshot for 24 hours, 1 daily snapshot for 30 days?
- Is it possible to keep a higher number of snapshots on a "DR-system" with asynchronous replication?


In theory, LSFS should be slow on "uncached reads".. That should be solvable with all-flash-storage, but the LSFS-overhead makes this quite expensive. How do you solve this? Just an example:
7x 1.6 TB Flash
--> Raid 5 (brave with 7 disks...): 9,6 TB
--> Max LSFS-Size: 3,2 TB

And a last question:
Are there any plans to add the feature: "clone-snapshot" to get a read/write-version of it?


Regards,
KPS
Oleg(staff)
Staff
Posts: 568
Joined: Fri Nov 24, 2017 7:52 am

Wed Dec 19, 2018 10:52 am

Unfortunately, LSFS is not recommended option for the most of All-Flash scenarios.
You can find more information by following this link.
KPS
Posts: 51
Joined: Thu Apr 13, 2017 7:37 am

Wed Dec 19, 2018 2:13 pm

Hi!

So, there is no "starwind-solution" for All-Flash, if snapshots and aynchronous replication are needed - right?
Michael (staff)
Staff
Posts: 317
Joined: Thu Jul 21, 2016 10:16 am

Wed Dec 19, 2018 6:50 pm

Hello KPS,
LSFS was designed for spindle storage to coalesces small random writes into the stream of big sequential writes to get better performance. It can be used on some flash configurations, but as you mentioned it's expensive :)
So if you really need LSFS feature, you can configure it on All-Flash storage.
KPS
Posts: 51
Joined: Thu Apr 13, 2017 7:37 am

Wed Dec 19, 2018 7:05 pm

Hi Michael!

Thank you for your answer!
Can LSFS provide a good performance on flash? I did not find any informations about this?

Another question: does LSFS provide any "system" to do integrity checks like zfs or refs?
KPS
Posts: 51
Joined: Thu Apr 13, 2017 7:37 am

Thu Dec 20, 2018 2:29 pm

Hi!

It would be great, if you could give me some input on:

- Can there be "nearly" unlimited snapshots?
- Is there any possibility to see the space, that is consumed by every snapshot?
- Do you have retention-policies like: Keep 1 hourly snapshot for 24 hours, 1 daily snapshot for 30 days?
- Is it possible to keep a higher number of snapshots on a "DR-system" with asynchronous replication?
Michael (staff)
Staff
Posts: 317
Joined: Thu Jul 21, 2016 10:16 am

Fri Dec 21, 2018 2:45 pm

Hello!
LSFS is checking the data internally, but using another unique mechanism, which is not the same as zfs or refs.
You can read about working with snapshots here: https://www.starwindsoftware.com/resour ... -snapshots
There is no possibility to see the space, taken by the snapshots (unless you check it manually :) ) and there is no retention policies for LSFS, but you can use scripts here for snapshots automation: C:\Program Files\StarWind Software\StarWind\StarWindX\Samples\powershell
Also, you can read about Asynchronous replication here: https://www.starwindsoftware.com/resour ... r-recovery
During the Asynchronous replica creation, there is a possibility to set a maximum number of stored snapshots on the DR site (see step 10).
KPS
Posts: 51
Joined: Thu Apr 13, 2017 7:37 am

Mon Dec 24, 2018 9:20 am

Hi!

I did some performance-tests with LSFS and full-flash / NVMe, but results are not very good on random write.
It's sad, that there is no oterh snapshot/replication-possibility with starwind.
User avatar
anton (staff)
Site Admin
Posts: 4008
Joined: Fri Jun 18, 2004 12:03 am
Location: British Virgin Islands
Contact:

Mon Dec 24, 2018 10:07 am

Don't do LSFS with all-flash, it's designed for parity RAID (5,6+). For replication you'd better use application level replication (SQL Server, Veeam etc), it's where industry moves from LUN replication.
KPS wrote:Hi!

I did some performance-tests with LSFS and full-flash / NVMe, but results are not very good on random write.
It's sad, that there is no oterh snapshot/replication-possibility with starwind.
Regards,
Anton Kolomyeytsev

Chief Technology Officer & Chief Architect, StarWind Software

Image
KPS
Posts: 51
Joined: Thu Apr 13, 2017 7:37 am

Mon Dec 24, 2018 11:47 am

Hi anton!

At the moment, there is no replication solution except zerto (which I know). Veeam-replication 9.5 leads to an outage on every replication...

Everytime, Veeam is doing a snapshot or deleting the snapshot, 1-2 pings of the VM is lost and every UDP-packet, that is sent to the VM is lost.
--> 2 outages on each sync-job on each vm

Thats why i am looking for a storage, that can do the job.
User avatar
anton (staff)
Site Admin
Posts: 4008
Joined: Fri Jun 18, 2004 12:03 am
Location: British Virgin Islands
Contact:

Mon Dec 24, 2018 11:54 am

LSFS won't help you here... Sacrificing performance (dramatically) in favor of async replication is a bad idea. Can you use per-application replication? What's your workload?
KPS wrote:Hi anton!

At the moment, there is no replication solution except zerto (which I know). Veeam-replication 9.5 leads to an outage on every replication...

Everytime, Veeam is doing a snapshot or deleting the snapshot, 1-2 pings of the VM is lost and every UDP-packet, that is sent to the VM is lost.
--> 2 outages on each sync-job on each vm

Thats why i am looking for a storage, that can do the job.
Regards,
Anton Kolomyeytsev

Chief Technology Officer & Chief Architect, StarWind Software

Image
KPS
Posts: 51
Joined: Thu Apr 13, 2017 7:37 am

Mon Dec 24, 2018 12:16 pm

Hi!

My workload is a vSphere infrastructure.
I am using 4 procudtion-hosts and a DR-vSphere-node.

My goal is to replicate the data to the DR-location to be able to bring the VMs up, there, if needed.

I am testing some different solutions, but I hope to be able to stay with starwind (as I really like your product and your support!):

- LSFS -> has all the features, but bad performance with all-flash
- ZFS -> no good products with good support in Europe on the market, but full featureset with replication, snapshots, snapshot-clones, etc.)

Replication on VM-basis:
- Veeam 9.5: Outage on every replication
- Zerto: VERY expensive
- Veeam 10: should be out for months/years...
Michael (staff)
Staff
Posts: 317
Joined: Thu Jul 21, 2016 10:16 am

Sat Dec 29, 2018 4:11 am

Hello KPS,
Thank you for the provided description and feedback.
You could replicate data to the DR location with asynchronous replication, but in your case, I believe, it is better to go with Veeam B&R or any other solution to replicate VMs.
Post Reply