This cisco spec sheet on the fabric modules:
http://www.cisco.com/en/US/prod/collateral/switches/ps9441/ps9402/ps9512/Data_Sheet_C78-437760.html

Says the system does “46 Gbps per slot per fabric”.

You said above that it does “230Gb/slot”.

Is that 0 a typo?
Is the cisco product spec doing it’s normal marketing double counting?

I guess my amazement comes from remembering the days of connecting 7500 Series Routers to Mainframes with a faster Front-End Processor and then from the shift to Client-Server and SOA apps driving the most network-centric application architectures seen yet.

I don’t want to say that the network reigns supreme, but I would argue all day that a data center needs networks, servers, and storage- all working in concert together, in order to efficiently process workload.

Today- I’m pretty happy we have a good seat at the table, 3-4 years ago it was a different story. 3-4 years from now…

dg

Cooling capacity is the real issue when you get down to it. I have 230Gb/slot in the Nexus 7000 (to correct one errant post) and about 800w/slot. In today’s VM focused environment I would need about 64-128Gb/Socket and given the board sizings on the Nexus 7000, (roughly 17″x20″) I would hope to fit 2-4 sockets on it. Given the power densities of DDR3 today and the demands of a quad to octal-core chip per socket there is no way I could come close to powering it much less cooling it in the Nexus 7000 or Catalyst 6500.

I can assure everyone the Nexus 7000 and Catalyst 6500 will continue their development trajectory as NETWORK platforms, not as SERVER platforms. Although, suffice to say, we’ll continue to focus on delivering the best NETWORK platforms out there :)

The Linux blades offered today are purposed as appliances, usually with a mixture of x86 and NPU/ASIC data paths. None of them offer an open compute environment for development of apps. The bulk of these are used for things like Firewalls, Server Load Balancing, IDS, etc.

dg

“Not a Chance!

The slots in a Nexus 7000 are too valuable and high performance to waste on a simple server. At 230G/s, unless you are looking at putting 20 servers in a slot, this would be a complete waste of money. Cisco wants to make $40K per slot, not $3K for a blade server.”

There are other places in their portfolio where this may make sense but on the Nexus? Pass me what you’re smoking. That’s a bad call, Allan.

Why do you think Andy Bechtolsheim left Cisco to form an AMD blade company? (which was smartly bought by Sun!)

Frankly, I don’t trust Cisco to be able to execute on this type of product. They’re simply too big and politicial to pull it off.

Also the Nexus 7000 would be a waste of that platform.
Each slot on the Nexus 7000 can push 80Gb. You’d need at LEAST 8 machines per slot to make the backkplane useful. There would be little point as a customer in spending so much on a chassis and backplane to only run a few gigs per slot.

So I think you’re way off on a few points here, but it’s fun to talk about.

@Jay - thanks for the thoughts on IBM having better server expertise. And, as you imply, you have IGS there to help you out.

How does it happen? I argue that the recovery from bad network engineering has a clearer path to success. You cough a relatively bounded sum of money for new networking equipment, deploy it in a relatively bounded time, and fix the problem. Recovery from bad server engineering has a much less clear path to success. Because the success of the application has many connections between servers, VMs, memory, caches, disk storage, and networking, the recovery from bad server engineering is much less deterministic. It is harder to predict when you will recover, and if you have enough money to recover.

In the market you describe, IBM has the server engineering expertise, while Cisco and IBM both have the network engineering expertise. Having Cisco peddle blade servers won’t change that.

I’d go with IBM. If Over time Cisco builds branded enterprise server engineering expertise with a “Cisco Global Services” group (or something like that) that is well known, then I would give them the same advantage as IBM.