john1266
(John_iTBe)
1
Hi Guy’s
This saturday we Will update our Network
AD 6x 900gb Ent to dl380g8 to get 2.5 tb Raid 10
Lire this dl380 Will act as files server for hyper : questions :
hyperv over Smb3 or ISCSI Target ???
Who’s perform’s better ?
About 1tb Vm Will ne stores
Thx
John
12 Spice ups
Have you tested both to compare results?
What sort of connection are you using to the storage?
Why remote storage and not local?
3 Spice ups
john1266
(John_iTBe)
3
Hi Gary.
Thx for thoses questions:
ISCSI performed Well in our synology i never try in win 2k12r2 in Target mode.
Smb3 was just tested basicly and concept and speed Looking Nice with dedup
Unfortunatly actual speed Will hé limited to nic teaming 3 or 4 cards.
John
then your bottleneck is the networking.
iSCSI will out perform SMB3 but not by much and if you have more familiarity with iSCSI over SMB3 you’re better off using that as you’ve left it a bit late to thoroughly explore SMB3.
Local disks would be faster and less complicated.
3 Spice ups
john1266
(John_iTBe)
6
What do you mean by local dsk .
The idea is to have cluster in place
Storage
/ \
HOST A Host B
As it says. Disk that is LOCAL to the server, i.e. disk that isn’t on the end of an Ethernet cable.
2 Spice ups
Why do you want shared storage? What is the goal of the shared storage in the business? The usual first answer is for high-availability in case one of the hosts goes down. But that single shared storage is the exact opposite of high-availability. If the shared storage goes down then everything is down.
Using two hosts each with independent local disk storage allows you to eliminate any shared dependencies in order to achieve true high availability. Keeping the guest virtual machines up-to-date across both hosts would involve replication (either synchronously with something like Starwind Virtual SAN or VMware vSAN, or asynchronously with something like Veeam, Zerto, Doubletake, Microsoft’s own Hyper-V Replication, VMware’s own vSphere Replication, etc.).
Search around the Spiceworks community for more information on this as it is discussed constantly and there is a tremendous wealth of information and analysis.
@Veeam_Software @StarWind
4 Spice ups
SMB3 is still Point-to-Point while iSCSI (and FC) MPIO is one-to-many. So if you have many I/O nodes and multiple paths on a light workload iSCSI would be faster and effectively all nodes would participate in requests processing while SMB3 client will talk to just one node. On a heavy workload when many clients “talk” to many SMB3 shares it does not matter.
Upcoming Storage Spaces Shared Nothing situation would be again bad for SMB3, reason is file level protocol has no clue where actual data resides so client can easily start talking to server with no data on a local disks to extra I/O over network would happen, say probability to have no “parasite” I/O would be P = N / M where N is a replication factor (2 or 3) and M is number of SoFS nodes in a cluster. FC and iSCSI have custom DSMs for that and always know what I/O node or virtual controller to target to get data w/o request re-route 
In general it’s incorrect to compare SMB3 Vs. iSCSI as you need to compare EXACT implementations. If you’ll compare MSFT non-F/O SMB3 Vs MSFT iSCSI target then use SMB3. Performance would be the same but SMB3 is much easier to use
2 Spice ups
I have never used SMB3 for Hyper-V. Can you do it with multiple nodes and local storage? Or do you need some kind of shared storage like always (physical or virtual SAN)?
If you have local storage, why would you use SMB3 when you go direct to the drive?
1 Spice up
DoctorDNS
(DoctorDNS)
12
I would make sure I tested both before making my mind up.
Having said that, a clustered, scale-out file server looks a pretty formidable product. It’s cheap and with some sensible network architecture should run OK And you just need Ethernet.
Of course, you can go 10GB ehternet, or for extreme speed, Infiniband. Both would minimise any speed issues.
Definately ook at SMB 3 in Server 2012, and if you want to get a bit over-excited, look at what is coming in Server vNext!
1 Spice up
…probably because SMB3 would be still used as part of CSV even if doing “local” storage?
1 Spice up
You don’t need shared storage for SMB3 (if you don’t want to run it fault-tolerant of course). If you want FT you’ll have to put CA SMB3 shares on top of FT storage. Physical or Virtual. Upcoming version of Windows does not change anything except (finally! thank you very much!) dropping a requirement in a shared DAS for Storage Spaces (with Clustering).
1 Spice up
You should be using local storage (disks in the server) for a small deployment. SMB3 and iSCSI should not even come up.
2 Spice ups
by implementing SMB3, it would make it shared storage. SMB3 is just windows shares.
2 Spice ups
I’m glad you said that Scott. I was confused why someone above said SMB3 was point to point and iSCSI was multi-point. My understanding was exactly the opposite. The windows implementation of iSCSI (low end) only allows one server to connect to the storage.
Hello everyone! Thank you for referencing StarWind Virtual SAN™ !
@John, just want to confirm that in our opinion, it will be much more efficient if the local storage of the servers will serve as the shared storage, and that is something that StarWind definitely can do. This is how it will look like:
If you are interested I may offer you the live demonstration of our product - just PM me and we`ll scheule it for you.
2 Spice ups
You’re confused. Windows built-in iSCSI target can do MPIO just fine. See:
Windows iSCSI Target MPIO Configuration
Quite soon we’ll be posting an extensive research on Windows MPIO Vs LACP and it will show (as a side-effect) how Windows does scale performance with MPIO paths increased.
john1266
(John_iTBe)
20
Hi Guy’s,
Thx for all differents comment’s and point of view.
We discovered this Week-end during our migration SMB3 is powerfull but it’s not painless.
Finaly we tested Starwind San, and honestly it was easier to use and directly accessible via our cluster than SMB3 where we still stuck with Sharing security.
Let’s see with the time how it’s work
John
1 Spice up
