However, your other points just don’t match my experience. Many of them are somewhat subjective, so it is difficult to comment definitively, but on a point of fact, the buttons at the bottom of the screen are illuminated, so perhaps your phone is just faulty.

You’re right that not all applications will benefit from 1333MHz vs. 1066, but some certainly do. If it made no difference, Intel would not have spec’d the high-end procs to run the memory bus speed at higher clock rates, right?

Not only can HP not achieve 1333MHz in any configuration beyond 144GB, their half-slot BL servers (including, but not limited to the 460c, 490c, 465c, 495c) can achieve a maximum memory density of 192GB and all of it will run at 1066MHz. Those extra dimm slots don’t help you in the 490c because it exceeds the supported rank count when using 16GB dimms so you are limited to the normal 12 slots. This also is only possible using 16GB dimms which are very expensive at today’s prices. Cisco can achieve 384GB running at 1333MHz with standard 8GB dimms. Cheap RAM +Fast RAM=unbeatable combination. The difference is that Cisco creates ASICs for their products. HP buys them. Even in the flagship Virtual Connect line, you are looking at switching ASIC created and maintained by Qlogic (FCoE) and Broadcom (Etthernet). See http://itbrandpulse.com/Documents/Industry2010004%20-%20QLogic%20Enters%20Ethernet%20Switch%20Market.pdf

HP is able to get higher memory densities in larger footprint blades. The new BL620c(512MB) and the BL680c (1TB). However, the footprint of these servers is so large (1/4 chassis and 1/2 chassis, respectively) that your ROI would suffer greatly. You’d also be paying for 32 switch ports on Virtual Connect (or anything else you put in the c7000) for just four (BL620c) or two (BL680c) servers. You cannot scale a solution like that.

HP’s best core count per chassis is the BL620c. It gives you two 8-core Nehalem-EX CPU’s and 8 blades per chassis. That’s 128 cores in a 10U form factor – or 512 cores per rack in a max config. Compare that to the Cisco UCS B230 at 16 cores in a half-slot blade (same size as a BL460c) giving you 128 cores per 6U – or 896 cores per rack in a max config. That’s quite a density difference. If you are thinking that HP is providing more memory than the B230, you’re right….for now.

Since you said yourself that you have not seen UCS, it would be unfair to truly compare the two solutions (HP vs Cisco). But I can tell you this – to install HP’s SIM product with all the add-ons requires 27 services. The reason it requires 27 services is because all the add-ons were different programs written by different companies acquired through the years. This is also why it is so difficult to get an overall “OK” status for just a single chassis in SIM. If you use SIM, you know what I’m talking about. It’s a monster to install and maintain – and it has zero redundancy. If SIM goes down, it has no recovery mechanism. Cisco UCS requires zero services on a server and run entirely on the Fabric Interconnects in the dat center – and it’s redundant.

For me to buy into your argument that companies with a large IBM of HP investment would not look at alternatives means one of two things is true. HP and IBM are guilty of the FUD they sling about Cisco’s “Vendor Lock-in” OR the company is not interested in saving money. To buy into HP and IBM’s method of buying blades, you have to believe that for every 16 servers (maximum) you should deploy at least 2 switches (and in most cases today, 4 switches). Imagine what the datacenter would look like if all those access switches weren’t hidden on the chassis. The network team would be fired for a design of hundreds of access layer switches. Why do that when you don;t have to? Cisco UCS can provide all the redundant switching you need for 160 servers in just 2U of space. It’s no secret that we’re capable of doubling that when we complete some additional scalability QA on the 2U 6140 interconnects (4U for redundancy). We’re also experiencing 550% Q/Q growth in UCS. We started at 900 customers in our first report, then 1700 the next qtr and 2800 customers the next. We’re growing at an explosive pace and have no signs of slowing down.

On the network teaming subject, we use the FEX architecture in the chassis along with the fabric interconnects to provide network redundancy. So you only need to assign a single NIC to a server and if the Fabric Interconnect(FI) that it’s attached to goes down, UCS will automatically re-route that NIC to the remaining FI without the OS having any knowledge of the fact that it happened (it will never show link-down). Proliants have great teaming software – it works across Intel and Broadcom adapters in the same UI. But it requires software which means it requires maintenance on every single server out there. It also only works on Windows. So if you use Linux, who’s problem is it if the teaming/bonding fails? Is it a Linux disto issue or a NIC driver problem? Proliant solutions also mean exposing additional NICs to the OS to provide redundancy. I need just a single NIC exposed to the OS on Cisco UCS. you can read more on it here: http://bradhedlund.com/2010/09/23/cisco-ucs-fabric-failover/

Don’t you find it odd that you say we are a year ahead of the other blade vendors in the network space? HP has been producing Virtual Connect since 2006 and Cisco is ahead of them in just 18 months of being in this business? That’s your time estimate, not mine – having been inside each company’s Blade divisions, I can assure you that HP is nowhere close to catching the feature set of UCS. Even if it were true, in 1 year HP will be where Cisco UCS is today. Not where Cisco UCS will be in the same year. We are 100% committed to success in this industry.

I cannot agree with you on getting rid of the 6500 and 7000 switch line. In this line of thinking, HP just bought 3Com and H3C for no reason and should instead look to stick everything in the c7000. This will never happen in either world and there are a thousand reasons why.

Lastly, you started your article by saying that UCs would “try to be HPs BL” this year. I hope you now see that we have no intention of being anything like BL (outside of being a market leader). As for “no major company” buying UCS, I guess ING, Cable and Wireless, Savvis, Sunguard, Telstra, CSC, Logicalis, Tech Data, Intel, Netapp aren’t major companies. You can see a whole lot more of them here: http://www.mseanmcgee.com/2010/10/cisco-ucs-references/

So now I get to sit back and see how you do on your predictions

Thanks for writing up your comment, this is good stuff. Obviously I don’t have all the answers but let me address some things from my point of view:

* 384GB RAM @ 1333Mhz is indeed no small feat. HP has either the speed or the capacity but not both. now the question is how much of a differentiator is this in a corporate environment? Arguably if your application is so sensitive that it requires 1333Mhz instead of the 1066Mhz you could get a dedicated platform just for that application of you might consider fine tuning the application. Again, it’s an advantage but if it comes at a premium people might not buy it if they don’t specifically need it.
* 256GB RAM in half slot blade, I think HP supports similar densities in the G7 products. The BL465 supports this in a 16servers/10U design.
* NIC teaming, for FT – I will have to read up to understand how this is different from the ProLiant offer, maybe you can give some pointers?

From a management point of view I am yet to see Cisco’s tool. It may be brilliant but for sure it has to go against IBM’s and HP’s tools which have a lot of development behind them and more importantly, admins have been working with them for years.

You could argue that a 20-some year old, fresh out of university will learn it in hours but enterprises are worried about the staff that has been working in the field for long years which are a lot more resistant to changes. Re-training those can be a show-stopper.

I will look into the other features that you mention, some look interesting.

Obviously Cisco has a lot of experience in the networking field and I see the UCS being one generation ahead of the other vendors (HP is investing heavily in the VirtualConnect and Flex-10 which probably at the next iteration will be a worthy competitor), on the other hand the typical separation between server and network people that you see in most enterprises might not be doing Cisco any justice. Could it be that the server guys you are selling to will not worry about the advanced network features brought by the platform because that’s something the network team needs to worry about? I’m just speculating here.

Let me clarify a bit on the comment regarding switches in blade chassis: My intention was not to suggest to Cisco to take a me-too approach and put switches in the back of the chassis like HP does. No, sir!

I want to see the obnoxious 6500 side-breathing monster phased out. Let the 7600 go the same way and don’t forget to let the 4500 go also. Now come up with a chassis that can take servers and line cards and can become either a server chassis or switch chassis as needed. Maybe also allow some mixing although there are few scenarios that require that (maybe a branch office?). This would allow companies to stock ONE chassis type instead of four (3 switch platforms, one server platform).

Not sure if that makes sense for everybody but if you are concerned about training your staff and you have to stock everything on site for security and availability reasons than thins kind of consolidation makes a lot of sense.

Why would be want to put switches into the blade server chassis? That’s being done by HP, IBM, and Dell. It seems you are expecting Cisco to produce “blade servers” as they have been produced for the last 10 years. We wouldn’t be in the server business if we were here to do a “me too” on blades. We are creating what blade servers SHOULD look like – not what they have been in the legacy blade space. There are way too many unnecessary layer 2 decision points in an IBM/HP/Dell chassis design. For a VM on one chassis to talk to a VM on another chassis, it has to pass through 5 L2 decision points (including vswitches). With Cisco UCS, I can do the same thing in a SINGLE L2 hop. No other server vendor in the world can do that. And we’re just getting started….

December 30, 2010 By Nancy Weil | IDG News Service We were wrong — so far — that Carol Bartz would be ousted as Yahoo CEO by the end of this year, but we were right that Apple’s tablet, whose name was not known at th……