Cisco Blogs

Building a Better Blade Server – Network Style….

March 12, 2009 - 27 Comments

I’ve been asked a lot about why would Cisco build a blade server recently. I don’t know where these rumors and speculation come from. But notwithstanding I thought I would write out the answers I usually give. 1) Cisco is NOT building a blade server. What we are doing with Unified Computing is so much more than just a blade server that even using terms like blade server in the same sentence with Unified Computing doesn’t do the architecture justice. The problem is that with something that is this big in scope and scale, people often have a hard time wrapping their heads around it and conceptualizing it. Let’s face it though – in a couple years it may be the way people look at IT and they look back at the piece-meal, services intensive, high cost approaches they took from 1990-2010 and wonder why no one ever did Unified Computing earlier.

2) Ever plug a server in, turn it on, and watch it? It just sits there. Occasionally a light may blink – that is usually the status indicator on the network interface card. Why? because networks TRY to do something and their default operating behavior is to be interoperable and connected. The server, now plugged in, is sitting there, idling, consuming about 40% of its peak power, draining kilowatt hours, costing you $$$, and accomplishing nothing – unless you needed a marginally efficient space heater. Ever plug a router in? Or a switch? With 1-2 commands your router can join the network, advertise itself, determine what is connected to it, and start creating alternate communications paths, increasing the bandwidth and performance of the system as a whole. Networked devices communicate with open standard interfaces, work together, even across vendors, and start creating some business value out-of-the-box. I am sure a server vendor will come back to you with some answer involving a multi-million dollar management system that they want you to have, and then multi-million dollar services contracts and customization support you need to make it work and show how that can make the server be sort-of-like-a-router. 3) R&D spend. Some companies want to own everything – they want to sell you ink, services, toner, people, switches, servers, cables, people, storage, printers, monitors, proprietary Unix, people, gaming machines, photo printing, people, calculators and even paper (did I mention people…). In the data center, as in other areas of networking, we are focused on interoperability, standards, and going to market with a broad array of best-of-breed partnerships. Sure one company can try to own all of it, and everyone has some growth aspirations – hard to fault that or I’d be calling the kettle black, but if you look at the aggregate R&D and services delivery capabilities of Cisco and our Partners in this space – it is far larger than any monolithic and myopic company. 4) Innovate. Innovate. Innovate. At an event two years ago an analyst mentioned to me that ‘Blade Servers are the biggest innovation in the server market in a decade!’. I was shocked, astounded, and somewhat stupefied. I asked back, after a moment of pause- ‘If rotating a server 90 degrees on its side is the biggest innovation in the server market in the last ten years don’t you think the server market has a problem?’ (we did that with switching 10+ years earlier). I went on for the next two years asking many people the same question – I have heard back ILO boards, Virtualization, Multi-Core, etc. I remember using lights-out management boards over 10 years ago, so that sort of rules them out; Virtualization was not delivered on x86 by a server vendor, but instead by a software company; and Multi-Core CPUs were developed by the CPU vendors- Intel and AMD. So what have the server vendors done for their customers in the past 5-10 years? Where has all of this R&D spend gone? I’ve seen server margins compact, and server companies staff thousands to make up the margin with lucrative services. In fact some server vendors have started moving to integrate more tightly with the network- why? Because every time network capabilities have increased two things have happened: parallel networks have consolidated and servers have disaggregated. (Don’t believe me? Look at IP Telephony and FibreChannel and FCoE as examples). Why not make a server that creates customer value. Then you can solve real business problems. Then you can return shareholder value. dg Follow me on Twitter @ dgourlay

In an effort to keep conversations fresh, Cisco Blogs closes comments after 60 days. Please visit the Cisco Blogs hub page for the latest content.


  1. SEO Company India: For Best SEO Services, PPC Services, Link Building Services – Contact SEO Professional India. Best Ethical SEO Search Engine Optimization & Internet Marketing Services Experts, White Hat SEO Specialists Company in India

  2. I watched the webcast yesterday, and indeed, Cisco IS building a blade server (contrary to Doug’s item #1 above). The webcast did not really get into details, they talked primarily about how all the wonderful partners are going to join with Cisco to solve world hunger, er, I mean, solve all the problems in the datacenter.From scouring, I found the following info (and hopefully someone from Cisco can correct me if I’m wrong please):6U blade chassis with 4 server slots. It that can hold 8 half-slot blades or 4 full-slot blades.* Observation: This is a pretty small chassis. I can get 14 blades with IBM BladeCenter and 16 blades with HP BladeSystem.Chassis has 8 fans and 4 power supplies (no mention of their redundancy model).* Observation: Lots more fans per server than either IBM or HP, more power supplies per server as well.Server blades can only handle up to 4 10GbE ports each, which probably means that the half-slot blades only handle 2 10GbE ports each.* Observation: This is on par with IBM, but a lot less connectivity than with HP.The blade chassis only holds two fabric extenders””, so it looks like there’s no chance to do 1 Gigabit Ethernet, Fibre Channel, Infiniband, or anything *other* than 10GbE.The blade chassis can *only* connect to Cisco’s external 10GbE switch, which comes in 20 port and 40 port models.I haven’t seen any pricing yet, but if I need a small blade chassis with just 6-8 servers, I’m certain that an HP or IBM solution will be much more cost effective.One thing did pique my interest: Their management software (in the UCS switches) works to combine 320 blades (40 blade chassis) into a single “”system””. HP says their Virtual Connect system can handle up to 2400 blades as a single cloud, but you have to pay extra for that.”

  3. Omar-Thanks for the Cisco view of my assertions, but you were missing my point about idle network switches. Surely, if no ports on a switch are connected, it wouldn’t make any sense whatsoever to have it powered up — except to function as a nice foot warmer.My point was that switch ports are idle when they are not transmitting packets, which can be a lot of the time. (Of course, I’m talking about the numerous edge switches in a network that are connected to end nodes.) Yet these switches are drawing the same amount of power 24×7 — they do not have low power”” modes like CPUs. This wasted energy can really add up in a datacenter, which is why the IEEE’s Energy Efficient Ethernet study group was created. I hope that someday we will have ways to power down switch ports when they are not carrying traffic.Your comparison of a CFP to a CCIE is quite interesting. Financial planning is very complex – a lot like learning IOS. Which was exactly my point — you don’t want just anyone configuring your Cisco switch, because they could easily make a mess of it. Hence, the numerous certifications.Do you have plans for a Cisco Blade Server Administrator (CBSA) course? It would be another possible revenue stream…”

  4. The coolest part is probably that setting up PHP services for integration with ActionScript (using AMF) is now a pretty automated process. In FB4, there’s a wizard that will set up all your base code, includes, and file/directory structure with a few clicks. Obviously, for extremely custom implementation

  5. You are perceptive that the management is key to the UCS—both the BMC options and some of the other approaches that are possible for heterogenous environments. More detail on this is forthcoming!!

  6. i am not an expert on server architecture. but interested to know, which one is better… Grid server or blade server ?

  7. hi Douglas Gourlay !I am really shocked by the disdain you express for server vendors too.

  8. Thanks for the Cisco view of my assertions

  9. Sure, blade servers are very much useful than conventional rack servers. They are needed to save place.

  10. I always look forward to the innovation of Cisco servers.

  11. Ignacio:Actually, no, we did not release a blade switch, we released a system that integrates a number of resources, including compute and I/O into one system that is managed holistically–which was Doug’s point. The UCS is sold as a system, you cannot simply by blades and a chassis and wire it up some other way.You are perceptive that the management is key to the UCS–both the BMC options and some of the other approaches that are possible for heterogenous environments. More detail on this is forthcoming, so stay tuned.Omar

  12. Ignacio:I appreciate the your passionate defense of server vendors. Unfortunately, a number of your assertions don’t withstand scrutiny. Taking multiple individual rack-mountable devices, centralizing common components like power and cooling, unifying management and allowing customers to add capacity in a granular manner is is great idea–in fact, SynOptics delivered this in the networking realm in 1989–it was over a decade later, in 2001, when RLX (acquired by HP in 2005) delivered the first modern blade server–which I believe is Doug’s point.As far as power goes–the idea of having a network switch idling doesn’t make a whole lot of sense–if you have a switch not connected to anything, then unplug it. In fact innovations like virtual device contexts let you virtualize a physical switch to maximize utilization and derive the maximum work for the power being spent and consolidate multiple physical switches into logical switches on single physical switch. Of course you can do this on a server these days, but that is courtesy of companies like VMware, Microsoft and Xen.I will argue that career certifications or the number of certs available have nothing to do with product complexity. They do, in fact, serve the needs of both networking professionals and employers. For employers, the certs give them them a standard to assess a candidate’s expertise, which help them make good hiring choices. If you are looking for a money manager, you could hire someone with a Certified Financial Planner certification or not–they might be equally good, but, if it is your money, what are your really going to do? From an network professional perspective, we used to only have the CCIE cert, but not everyone has the time, resources or need to be a CCIE, so we have a more granular approach with lets professionals move up the ranks at their own pace, to an end-point that they define, with the ability to validate their current level of expertise. Much like moving from an Associate degree, to a Bachelor’s, to a Master’s to a Doctorate.As far as whether Doug can configure a device from the IOS CLI–I am fairly confident he can, even lacking any career certification, but perhaps he will provide the authoritative answer to that one. :)Regards,OmarCisco Systems

  13. Brocade CEO Mike Klyako comments on this announcement here:

  14. Brocade CEO Mike Klyako comments on this announcement here:

  15. Your earlier comment that why hasn’t anybody thought of this years ago humored me. We did think of it 25 years ago, they called it the mainframe!! Yes it was a closed system, no it didn’t have the same connectvity or portability as x86 and yes the user was never independant which unltimley explains the mainframe’s demise but at the end of the day this is a mainframe play all over again. It’s good be be older and still in IT, sort of like been to the bar and got the T-shrit!!.-mc

  16. Ohh… well, Let me see…, Now that all other avenues are getting narrower and cash is less. Cisco will start selling servers…, Linux servers. I’m just curios how many versions and flavors of Linux there will be for those servers and how well they will work with the rest of the world…??? bare in mind the variety of IOS images and trouble shooting of it on command line level… when other players enter the game. Sure it will be open source… You don’t buy just Cisco… do ya?My personal opinion is…, Cisco is desperate after the market for network hardware is not predominantly theirs anymore, -plus their spending spree in the last several years, -plus the lack of creativity… (buying competitors is not creativity)…, -plus the high prices…, Soon enough Cisco will start competing with Telecoms and sell traffic to general public… That ought to hurt… but then what do you expect from a company that has it’s nose stuck in the clouds. Ignacio…, there will be course in 4 parts… it will cost you and arm and it will not guaranty you employment.

  17. Over the last few decades, there has been a constant innovation’s race between connectivity and computing disciplines (note: computing = server HW + SW). Computing heads were ahead for the past few years, with server virtualization and workload mobility, but connectivity heads are fighting back!In reality, it is not either or – we need both. Looks like we are entering the era of connected computing””…More connectivity vs computing details at”

  18. Cue up a little Buffett- ‘Come Monday, it’ll be allright, Come Monday….’I wonder how fun this thread will get in the next week or so…

  19. Monday, monday, monday…’twill be a fun day.California Dreamin’./Hoff

  20. I have to agree with Rodos, that when the NDA is lifted this will be a great discussion. Until then there are a few of us out there that would love to participate in the discussion, but have surrendered our keyboards to the legal department…–Colin

  21. I assure anyone who doubts me- that I can still configure an IOS or NX-OS based switch or router. I am nowhere near as skilled as one of today’s CCIEs- who have it far harder today that when I took that test back in 1999. However, I think I could hold my own on most configuration types for L2/L3 systems.dgoh, coincidentally – CCDA/CCDP #001. (been a while… a long while…)

  22. While I agree that we expect a server to be underutilized out of the box, before virtualization. That is the point of all the different virtualization companies out there, so to say that a server just chews power and a switch does not is becoming more and more antiquated. I partially agree that a blade is a blade, Sun is starting to prove/move to the hardware is a commodity. So to say that this is so different than any other blade is not necessarily true, unless you want to ignore the other advances in technology. VMWare and others are working more and more to move the virtualization to the hardware to make it a commodity. But I do have to say that setting up a Cisco switch despite all the certs is not that hard, I don’t know about 2 commands unless your not securing anything, however Cisco IOS is not that difficult. One last side thought on my random run here is I remember when HP was touting that everyone would be using RAID ram and hot swap ram, after they bought DEC. Now honestly how many use that? My only point here is that sometimes R&D finds cool stuff, just no market.

  23. Doug, how did you get your NDA lifted? 🙂 I don’t think serious conversations about this can really occur until something exists that people can discuss, rather allowed too. Flirting around the edges, whilst creating some expectation, does little. Looking forward to when this change occurs, I think it will be a good discussion.

  24. OUCH: The server, now plugged in, is sitting there, idling, consuming about 40% of its peak power, draining kilowatt hours, costing you $$$, and accomplishing nothing – unless you needed a marginally efficient space heater.”” SO how do you pry off the layer of server-huggers that are clinging to them for warmth?”

  25. Wow. I am really shocked by the disdain you express for server vendors. My mother would tell me that you must be very jealous. monolithic and myopic company””?Server vendors have been trying to make things simpler for their customers, and to say that a blade is just a server rotated 90 degrees is missing the whole point.You don’t like it because idle servers consume 40% of their peak power — yet any Cisco switch is probably consuming 90% or more of peak when idle. Also, you can power down a server when it’s not needed and bring it back online upon demand, but you can’t do the same with a switch – so where’s the innovation in switching?People? As if Cisco never tries to sell people? Have you, Douglas, ever really tried to configure a Cisco switch out of the box using Cisco IOS commands? It’s really not possible unless you’ve achieved one of Cisco’s numerous certifications – this is your real “”lock”” on the networking market. Check out the long list of Cisco certifications which indicates just how difficult their hardware can be to set up (CCNA, CCDP, CCDA, CCNP, CCIP, etc.) think the real truth is that Cisco is losing control in the datacenter and California is your attempt to try to regain control (in order to keep selling your high margin switches and routers.) You’re looking to increase CSCO shareholder value — not your customer’s shareholder value.”

  26. You can call it Unified Computing or you can call it a solution, but in reality it’s still a blade server. Now, it may be a very innovative blade server that is integrated with a lot of very cool technologies, but in the end it is still a blade server. Trying to call it something else is just being disingenuous.

  27. Certainly blade servers are the most innovative development in the last 10 years, but it’s not the orientation of the motherboard. It’s the shared resources on the back-end: integrated switches, shared HBAs, HA power, shared storage, etc. But this is not to say it’s all that innovative in comparison to, say, server virtualization!I am struck by the differing visions put forth by Cisco, Sun, EMC/VMware, and Microsoft (to name four)! Hold on everyone, this is going to get bumpy!