2012 std r2 w/ hyper v clustering and smb 3.2, need help.

[ElementalWindX]

I have 3 2012 std (all 2012 on this network) servers setup as such:

4 nics going to a virtual server on 2 vm hosts
8 nics going to a sofs server
All clustered together to host ad/dhcp/dns/filesharing

cisco switches, all lacp setup and verified.

Tried using Intel's NIC teaming (these are the top of the line 1GB intel nics)
Tried using the OS's SMB teaming.

Either way when I transfer files from my SoFS share to inside a VM, I get crap speeds. 30MB-50MB/sec, instead of 110+MB/sec.
At first I was getting speeds in the KB range until I changed the (Get-Cluster).BlockCacheSize to something like 4GB, so that helped quite a bit.

I can transfer a file to the node/host from the sofs file share and max out at 113MB/sec just fine.

I cannot seem to figure out what is causing this.

I've turned off RSS, and Queueing while under SMB team and Intel team and got nearly the same results.

Does anyone know what could cause this?

Also can someone tell me this, with the newest SMB 3.2 and multiple nics in LACP with SMB multichannel, am I able to get a 2+2=4 effect with the bandwidth? Or is this just some nifty new feature to prevent downtime when a nic suddenly dies?

 

[YeOldeStonecat]

That "virtual machine queue" is what I've disabled to gain throughput back..did you also disable that under each physical NICs properties?

 

[ElementalWindX]

That "virtual machine queue" is what I've disabled to gain throughput back..did you also disable that under each physical NICs properties?

I disabled it all on the intel teams method and noticed some minor gains.

When I was teamed thru the os I just disabled it thru the multiplex adaptors which I imagined forced the settings onto each adapter.

 

[YeOldeStonecat]

When I was teamed thru the os I just disabled it thru the multiplex adaptors which I imagined forced the settings onto each adapter.

I'm not sure if it will or not....mebbe check and kill it individually on each adapter.

 

[ElementalWindX]

Heres a question for you.

Say I have 4 vm's on 2 nodes all located on the same 2TB CSV file share.


If I were to break that 2tb csv share down to 4 smaller shares and host each VM on each share, would I see any performance gains over just hosting all 4 on the same big share?

 

[YeOldeStonecat]

Same spindle..shouldn't be a difference. It's equal to partitioning a large HDD...speeds of the HDD is the same.

 

[ElementalWindX]

Same spindle..shouldn't be a difference. It's equal to partitioning a large HDD...speeds of the HDD is the same.

I figured as much.

I ordered 5 more 3TB constellations to use for sql so hopefully that will speed that up at the very least.

 

[ElementalWindX]

From your experiences would you consider this to be okay benchmarks for a sql virtual server?


Maxes out at 140MB/sec read 115MB/sec write.


I was hoping for better performance.

Just to give you an idea of specs:

The csv server is running a e5-2620 xeon 2.1ghz w/ 32gb memory and 8 nics, server 2012 r2, 7 Seagate Constellation ES.3 ST1000NM0023 1TB 7200 RPM 128MB Cache SAS 6Gb/s drives.
OS drives are 2 SSD in raid 1

the vms are running e5-2650 v2 2.6ghz w/ 64GB memory and 8 nics, server 2012 r2, and 4 ssd drives in raid 10.


I figured I would have gotten more out of this setup.