Unable to allocate more than 10Gb/s bandwith in server profile

[JSmid]

IHi,

I'm currently testing HP Oneview for our Test environment and I'm having an issue regarding server profiles and bandwith allocation. I'm using an active/active configuration. Currently, my server profile has 6 connections. Connection 1 and 3 share VC 1 X1 and Connection 5 is a tunnel on VC1 X2. When I try to edit the requested bandwith to 4GB/s for Connection 1, 6 Gb/S for connection 3 and 5 Gb/S for connection 5 I receive an error that the total requested bandith of 15 Gb/S from connections [1,3,5] exceeds physical link capacity of 10Gb/s. Is it supposed to work in this manner or am I missing a configuration setting.

I'm using the HP VC Flex-10/10D modules.

[ChrisLynch]

Hello JSmid, and welcome to the HP OneView Community Forums.

What you are attempting to do is not possible within HP OneView today.  The associated downlink port from the VC Flex10/10D module has a maximimum throughput of 10Gb.  In your example, Connections 1, 3 and 5 matp to FlexLOM1,  Port 1, FlexNIC A, B and C.  Each FlexNIC can have a maximum of 10Gb assigned, but not all can be assigned the maximum, or a combination greater than 10Gb (or 20Gb in the case of our 20Gb adapters.)

This is different than what Virtual Connect Manager can do with the Min/Max Bandwidth allocation.  Do know we have this on our roadmap, but I cannot discuss releases or timeframe of when.

[JSmid]

Thank you, that explains it.

So if an adapter has a requested bandwith of 2Gb/s will it use up to 10 if it's available or is it locked in to 2?

[ChrisLynch]

The connection that maps to a specific FlexNIC will be set to the requested bandwidth, which would be 2Gb in your example.

[PatrickLong]

Chris is this still the case, that bandwidth per FlexNIC connection is static per the "requested bandwidth" setting you give it and cannot dynamically take advantage of unused bandwidth from another connection on the same FlexNIC?  i.e. a connection that is set at 2 GB requested bandwidth will never exceed 2 GB even if it is completely saturated while other connections are underutilized? 

Separately, I too have seen a 10GB cap enforced when trying to create a new profile template on my OV 2.00.x environment - is this enforcement a change from OneView 1.20.x to 2.0.x?  I have two OV environments that were initially set up *exactly* the same and one was upgraded to 2.00.07 while the other remains at 1.20.08 - in the 1.20 environment,  I absolutely have 2 sets of 4 connections (1GB, 5GB, 6GB, and 8GB) with Requested Bandwidth obviously totalling >10GB per physical connection on my dual-port 630FLB and these >10GB settings (in either a 'template' profile or an actual profile) are not disallowed, but in my 2.00.x environment I just noticed that the same set of 4 connections on the same 630FLB no longer has those 1/5/6/8 requested bandwidth numbers after the update - they are now set at 1, 3, 2, and 4, respectively - and those were NOT manually changed by me to the lower numbers.  Did the OV 1.20 >> 2.00 upgrade in some way "force" my prior >10GB aggregate connection settings for that physical FlexNIC to adjust lower until they fit within the 10 GB max bandwidth requirement that OV 2.00.x apparently introduced?  Or have I missed something?

[PatrickLong]

...also wanted to add that ALL of my Networks (both VLANs and FC) and Network Sets have reverted to the default Preferred bandwidth settings of 2.5 GB  (Max 10 GB).  Could the upgrade to 2.00.08 possibly have reset them all to defaults if the aggregate values in 1.20.x exceeded the 10 GB maximum bandwidth per CNA?

[ChrisLynch]

@PatrickLong, yes, it has always been this way.  HPE OneView doesn't allow oversubscription at this time, including the upcoming 3.0 release.

[PatrickLong]

Chris - thank you for your response, but  when you say "always been this way" do you mean that the aggregate bandwidth of connections was not permitted to exceed the physical uplink bandwidth in ALL versions 1.x-3.x of OneView?  What could possibly account for the 'Requested bandwidth' numbers in the attached screenshot, which in aggregate exceed the 10GB physical capacity of each port of the blade CNA?  Each of my blades to which this profile is applied has either a 630 FLB Emulex-based CNA or a 630FLB Broadcom-based CNA and no other connectivity. If I try to recreate this profile on my DR site OneView 2.00.07 using identical connections 'Requested Bandwidth' numbers (on identical hardware) it will not allow me to save the profile due to violating the 10GB limit - which validates what you're saying about it not being able to allocate aggregate connection bandwidth in excess of the physical capacity of the CNA hardware.  Not trying to be snarky but mustn't OneView have allowed this at one point? Otherwise how could this profile below have been saved at all, much less applied to 11 BL460c blades across two different c7000 chassis where the connection numbers show same as below for each blade's profile'?

OneView 1.20.08 allowing aggregate connections >10GB physical limit of hardware?

[ChrisLynch]

The 630flb is a 20Gb capable adapter (2x20Gb, not 2x10Gb).  What interconnects do you have in each environment?  If the interconnects attached to the sever in the screenshot are VC FlexFabric 20/40F8 modules, then that's why you are able to set those values.  If the other location with 2.00.08 installed has Flex-10, Flex-10/10D or FlexFabric modules, then they are limited to 10Gb, which your adapter is then limited to 10Gb (even though it is capable of 20Gb).

HPE OneView does not provide the ability to oversubscibe bandwidth like Virtual Connect Manager can.  We are looking to address that in a future release.

[PatrickLong]

Chris you were absolutely correct - it was NOT the speed of the CNA on the blades that was causing the restriction to 10GB, it was the VirtualConnect modules as you said.  The 1.20 OV environment has blade chassis' with HP VC FlexFabric-20/40 F8 Modules while the 2.00 OV environment has a chassis with HP VC FlexFabric 10Gb/24-Port Modules.  I was told all chassis were configured "identically" and although the blades are in fact identical with 20GB CNA's, I neglected to validate that the VC's were in fact "identical'.  Mea Culpa and thank you for pointing me in the right direction!