Virtual Connect Stacking - Mandatory??

[UselessUser]

Hi,

Just a follow up to my post here:

http://forums.itrc.hp.com/service/forums/questionanswer.do?threadId=1349793

(If anyone else wants to reply I would be appreciative)..

I have noticed the stacking options available with VC (Ethernet modules in 1,2,5,6,7,8), the problem is I am trying to work out if stacking cables are mandatory for VC to work or not??

Also I do not understand why you would want the ability to redirect a server NIC from one mezzanine interconnect another

(And if you did, why did they not design it like this so it could be done internally, and thus save us a lot of money having to buy modules in the relevant bay, only to then just pass the traffic back to another one!!)

HP really needs to provide better more accessible documentation for some of their kit, we cant all afford dedicated specialists or consultants...

[JKytsi]

Well ... HP has these "quickspecs" available for all products.

quote from VC quickspecs....

Stacking links are used to interconnect Virtual Connect Ethernet Modules when using more than one
enclosure (future) or when using more than two Virtual Connect Ethernet Modules in a single enclosure.
All Virtual Connect Modules within the VC Domain must be interconnected. Any combination of 1Gb
and 10Gb cables can be used to interconnect the Virtual Connect Modules; however the following
provides recommended configurations (note that a built-in 10Gb link is provided between modules in
horizontally adjacent bays). Supported cable lengths on 10Gb uplinks are 3 to 15 meters and supported
10Gb stacking links (connecting between two HP Virtual Connect Ethernet Modules) are 0.5 to 7 meters
Interconnecting the modules allows all Ethernet NICs on all server blades in the Virtual Connect domain
to have access to any Virtual Connect uplink port. By using these module-to-module links, a single pair
of uplinks can be used as the data center network connections for the entire Virtual Connect domain,
and allows any server blade to be connected to any Ethernet network.

[JKytsi]

...And if You are buying new HW in near future. I recommend the FLEX-10 VC modules and G6 blades --> no need (in most cases) to have 6 x ethernet modules in enclosure.

[Adrian Clint]

No you dont need to stack. Its an option to avoid going out of VC where you dont need to or do accomodate differing network configurations on different servers, or for plannin more granualr load balancing.

There are dozens of reasons why you might want to redirect a NIC out of a different virtual connect than the one its physically connected to. In your designs you have mentioned before I think its unlikely you would need to - thus probably why you dont see why you would need to.

There are stacking links designed already between horizontal VC modules.
Your other comment about design is already answered by Jarkko on Flex-10.

Dont try and squeeze VC into a config where it is not suitable or over specified ... it looks like you are heading that way.

[JKytsi]

But if you won't stack all the modules, can you manage them in one VC manager (which is running on bay1 or 2) ?

[JKytsi]

and it can't be the costs that prevents you from stacking. Because you can use the 1Gb copper ports for that too.

[UselessUser]

Hi,

Thanks for the replies.. already its conflicting... see where I am going here..

Post 1: All Virtual Connect Modules within the VC Domain must be interconnected. Any combination of 1Gb
and 10Gb cables can be used to interconnect the Virtual Connect Modules

Post 2: No you dont need to stack. Its an option to avoid going out of VC where you dont need to or do accomodate differing network configurations on different servers, or for plannin more granualr load balancing.

I am not trying to cause an argument here but just get the best solution...

I am particuarly interested in this statement though..

Dont try and squeeze VC into a config where it is not suitable or over specified ... it looks like you are heading that way.

May I ask what you mean by this?

Costs of stacking are not the issue... its more about ordering the kit basically.

Also do the BL460 G6's have FlexNIC's onboard or something?

[Adrian Clint]

Yup Jarkko is right about the managing all VC modules requiring stacking links for them to talk to each other. I forgot this. You can use a 1GB connection for this. No need for the 10GB links. Usually everyone stacks.. but I have seen a couple of instances where it was not done, cant remember reasons for config.

Yes all G6 servers have flex-10 compatible NICs.

As for trying to config VC where it may not be required. To me your requirement looks like you need 2x Cisco 3020 for production. 2x BnT Gbe2L2/3 for Vmotion and Service console and two pass-thrus, and two SAN switches.
What benefit do you think you will be getting by going to Virtual Connect? Fast profile moves of your VM hosts keeping the same MAC addresses? But you seem to have negated that ideal scenario for VC by moving the DMZs onto pass-thrus. Where the MAC address of the DMZ NIC will change when the VC profile moves.

[UselessUser]

Hi,

OK I am seeing this...

To me your requirement looks like you need 2x Cisco 3020 for production. 2x BnT Gbe2L2/3 for Vmotion and Service console and two pass-thrus, and two SAN switches

What is BnT? And why different switches for Prod and Vmotion etc...

I wanted Virtual Connect for the profile aspect, we have no restrictions on mac address for DMZ so the changing of this per VM host would not be a problem.

Is there a diagram somewhere of a topology layout if using Flex-10.

We are running BL460G1's at present but will be looking at BL460G6's very soon. So it might be prudent to take a look at Flex 10 as per both of your suggestions.

It looks like (correct me if I am wrong), the port mapping situation wont change, so onboard flex modules on the G6 will still go to IC's 1 and 2. Difference being of course, I could divide them up into 8 NIC's, all of which would come out of IC 1 and 2 (Saving me 4 IC module bays)..

[UselessUser]

Also to add... I elected for Pass-Through modules just to give a clean seperation between DMZ and Production networks...