topic Blade System Matrix and converged networks in BladeSystem - General

Blade System Matrix and converged networks

Justin Hannan — Wed, 26 Aug 2009 13:55:01 GMT

Hi

Does anyone know if the HP Blade Systems Virtual connect modules (VCe and VCfc) would work in a seamless fashion with the HP 2408 FCoE Converged Network Switch?

If so, does the HP Blade System Matrix support this setup? Essentially bringing it inline with the recent Cisco UCS systems...

ta

Re: Blade System Matrix and converged networks

Adrian Clint — Wed, 26 Aug 2009 14:37:25 GMT

I didn't see why it wouldn't. So I had a quick look at the details. It does not support Access Gateway, which means I would double check if it supports NPIV and I cant find a decent spec/doc that says if it will or wont.

Re: Blade System Matrix and converged networks

WFHC-WI — Wed, 26 Aug 2009 15:29:39 GMT

Hi Justin,

It sounds like you may need some adapters for the switch; the Virtual Connect Ethernet module ports are limited to CX4 connectors, not SFP+.

Other than that I don't see cause for concern.

good luck

Re: Blade System Matrix and converged networks

c3_1 — Thu, 27 Aug 2009 09:18:28 GMT

VC Flex-10 has SPF+

Re: Blade System Matrix and converged networks

WFHC-WI — Thu, 27 Aug 2009 14:07:11 GMT

That's correct... are you using VC-Enet 1/10GB modules or VC Flex-10?

Re: Blade System Matrix and converged networks

Justin Hannan — Thu, 27 Aug 2009 14:14:16 GMT

Thanks for the replies. I essentially looking at a data center refresh for the company. The HP Blade System using VC Flex-10 modules as it's likely the c7000 would be using BL495 G6 for a fairly large VMware farm.

I like the sound of the converged networks that the Cisco UCS offers so naturally am after seeing if the Blade System can do a similar thing.

My main task at the moment is to compare traditional rack mount vs blades vs cisco usc for compute power, energy savings, cable savings etc.

Eg. at the mo a c7000 with BL495G6 can provide me with 384 virtual CPUs (2 CPU x 6 Core x 16 Blades x 2 for vCPU), if that then amounted to 384 vm guests would a 10Gb flex-10 module be enough bandwidth for that many virtual servers.

Re: Blade System Matrix and converged networks

Adrian Clint — Thu, 27 Aug 2009 17:28:35 GMT

Well depending on how you configure the network you could have upto 2x (6*10Gb + 2x 1Gb) bandwidth usuable out of the enclosure
= 124Gb = 300Mb per CPU

The question is - is that enough for you?

And thats "upto" - you probably would not want to configure it that way. And it would be knife edge on the config with no room for a failing VC module.
So more like 100Mb

Re: Blade System Matrix and converged networks

Daniel Bowers — Thu, 27 Aug 2009 22:16:44 GMT

>>if the HP Blade Systems Virtual connect modules (VCe and VCfc) would work in a seamless fashion with the HP 2408 FCoE Converged Network Switch?

The uplinks coming out of VC aren't CEE, so while they would connect at 10GbE to the 2408, you wouldn't get CEE.

Quoting from HP's site:
>>When the CEE standard does emerge â and HP is helping to shape that standard today â you can be assured the Virtual Connect family of products will support it

I think Adrian's configuration is with one (or two) VC Flex-10 modules (am I right?) With more modules (and more mezz cards in the blades), you could boost that bandwidth.

BladeSystem Config: c7000, 6 x VC Flex-10, 16 x BL495 [each w/ 2 dual-port 10GbE mezz]

Uplink bandwidth from enclosure:
6 x (6 x 10Gb) = 360Gb (or 180Gb, redudant)
Using Justin's bandwidth-per-vCPU rule, that's 180Gb/384vm = 468Mb/vm

UCS Config: 5100, 2 x 2100 fabric extender, 8 x B200 servers

Uplink bandwidth:
2 x (4 x 10Gb) = 80Gb (40Gb redundant)

Using quad-core Xeon-based B200 M1 blades, you'd see
2 x 4 core x 8 blades x 2 for vCPU = 128Vcpus

40Gb/128vm = 312Mb/vm

BTW, the "convergence" of Matrix isn't converged *fabrics*, it's converged resource -- compute, storage, management, etc. At the data center level (especially a new data center -- lucky you, Justin!) whether the connection is FCoE or token-ring probably isn't as critical as whether the bandwidth/compute/storage/whatever can be added when needed, carved up and deployed, and done in the most optimal manner within your budget/SLA.