Comments on: HP Flex-10 versus Nexus 5000 & Nexus 1000V with 10GE passthrough

By: Steve

Steve — Wed, 25 Jan 2012 02:54:45 +0000

What you are comparing here is not even close to apples to apples. You are describing an HP solution with servers having 20Gb of IP bandwidth and 16Gb of storage bandwidth versus servers with 20Gb of combined traffic. Also in your QoS model and as you state, Cisco dedicates 50% for FCoE traffic. This leaves 5Gb for general VM, vMotion, and management traffic. You also mention that Qau is not required for FCoE while you are technically correct how do you scale without it lamost all storage manufactures are supporting single hop only without it. The silicon is just now hitting the street and products are still need. While I agree that HP has some work to do and that F10 is not for every site it is still a very good solution that fits a lot more sites than the Nexus solution.

By: Consultant

Consultant — Fri, 20 Jan 2012 17:46:43 +0000

I think of this situation as “grass is greener on other side”. If we set our mind set (as per suggested example) that server only has 3 NICs of 2Gb, 2Gb and 6Gb respectively, then things look simple. People (including myself) have burnt their hands by letting the things go out of the control i.e. automated management of bandwidth (or anything…). I am a consultant who manages ESX/Network and worked in several environment where educated customer had no issues coordinating between server/vmware/network teams.

Bottom line, 10Gb bandwidth on blade server with ESX is giant and AFAIK HP blades can go upto 80Gb network bandwidth with 2 onboard and 6 mezzanine NIC cards. Very very unlikely you will find a good case of not having enough bandwidth.

Finally, it’s all about three things… planning, planning and planning. In a planned environment we will never look at QOS a deal-breaker. And if we don’t have expertise to plan, we will always get into such mind boggling discussions.

By: Giograves

Giograves — Fri, 18 Nov 2011 04:39:28 +0000

HP/Cisco have collaborated and come out with a Nexus Interconnect for the c7000. Who would have thought a FEX sitting in an HP enclosure? That makes the prospect of a hybrid vendor solution much more enticing over 16×2 pass – through cabling.

By: Aaron Theodore

Aaron Theodore — Fri, 11 Nov 2011 02:23:17 +0000

Cisco and HP have now released a Cisco Fabric Extender for HP BladeSystem.

http://h18004.www1.hp.com/products/blades/components/ethernet/extender/
http://www.cisco.com/en/US/products/ps11975/

This means benefits of passthrough, without the extra cables

By: Giograves

Giograves — Wed, 02 Nov 2011 06:34:38 +0000

First, VMware ENT plus with dv-switches and network IO control (NIOC) play well with the “issue” of carving flexnic’s and losing max “burst” capability.

Second , I need to understand the math better where 16×2 10Gb pass through will actually save over an HP solution using FLEX-10 (ah hem now also with the FLEX-FABRIC module available).

10Gb ports aren’t exactly cheap. The X2 transceiver, the single/multi mode sc-lc cable, the 10Gb sfp…that’s between 3-4 grand per port alone.The whole point of HP’s solution is to save cabling. That has been their biggest selling point.

Frankly I cant understand using c7000 blades if you are of the Cisco fabric extending, dynamic bandwidth allocation type of mindset. Weird.

By: Jerry Gallagher

Jerry Gallagher — Wed, 19 Oct 2011 14:22:06 +0000

“No. Customers are not asking for a 1:1 ratio of server to core. There will be oversubscription at the Nexus 5000 just like there is oversubscription at the Flex-10.”
Customers are asking for 8 uplinks on an hp chassis connecting to 4 different switches in the Data Center fabric. We are asking for very limited over subscription to the core, have a second switch between the cores and the blade chassis is just redundant.

By: Michael Muller

Michael Muller — Mon, 26 Sep 2011 18:26:20 +0000

Hi,

I heard about virtualized switches. And I have to admit, that I really can´t see a real benefit. So our SANs are connected to our servers with 2x 24 port 10Gbe switches, stacked together, equipped with DAC cables where possible and some SFP+ optics, plus 5 year of manufacturer full service for ~32000,-€ incl. VAT.

Talking about QOS, of course each basic layer 3 switch has QOS features nowadays.

By: Brad Hedlund

Brad Hedlund — Wed, 14 Sep 2011 21:22:33 +0000

Kevin,
You don’t need transceivers. The Nexus 5K supports the same HP DAC cables as you would use with Flex-10. The HP DAC cable provides 10GE for both ends of the link at approx $100 street price. It’s insignificant.
Carve your Flex-10 up however you like in a way that makes sense for you. The point remains that you are providing much less than 10GE of available bandwidth to your virtual machines for I/O. With the Nexus setup, you don’t have to do that, because the bandwidth allocation works more intelligently.

Cheers,
Brad

By: kevin

kevin — Wed, 14 Sep 2011 20:07:14 +0000

you seemed to have forgotten transceivers. 10GB SFP= $750 list (approx) x 32 in the pass thru and 32 on the switch=48,000 x 4 chassis, HMM add another 192K to your little Nexus 5K config so that would be 140,000+192,000 = 322,000

and to be fair for flex 10 there are 16 external ports and you need modules in the 5K too so 32 x$750= 24k. and 4 chassis that is 96K + 168= 264K

SO you are right Ci$CO wins more money again

Flex 10 option would be$264,000
CIsco plus passthru would be $322,000

I gotta pay for this stuff, both you and HP gotta remember we do have budgets. and truthfully the 2 Chassis I have with 15 blades each, ( about 45 various VMs each) work fine with 8 cables coming off the chassis( 4 per flex 10) the blades talk to eachother on the Vnets without having to jump all over the network. and I assign only .500mb for console nic. above you say 2GB for console?? why that makes no sense.

By: Brad Hedlund

Brad Hedlund — Tue, 01 Mar 2011 00:56:47 +0000

Yes, the Nexus 1000V works with *any* upstream switch, even HP VC Flex 10. You should however question the QoS capabilities of the switch you are connecting to Nexus 1000V to insure your QoS policy is consistent throughout the network. The Catalyst 6500 has pretty good QoS, the HP Flex-10 has none.
Keep in mind that Nexus 1000V and HP Virtual Connect Flex-10 are not competing products. One is a physical switch (Flex 10), the other is the software switch within ESX (Nexus 1000V). They are not mutually exclusive because they each work in different parts of the overall architecture.

Make sense?