Today Cisco (NASDAQ: CSCO) will unveil its long-awaited server line, one of the building blocks that was still missing from their Unified Computing vision. This announcement is bound to shake up the IT industry in general, and the x86 server market in particular. Even though this market has several well established players like IBM, Dell, HP and to a lesser extent Sun, Cisco is determined to extend its already impressive datacenter footprint to this market as well.
What Cisco CEO John Chambers called “Datacenter 3.0” during his keynote at VMworld back in september 2007 has now been rebranded as “Unified Computing”. Different name, same concept.
Cisco emphasises that this is a big new concept, where the most important part is the (unified) network and the integrated management. We suppose the traditional server manufacturers beg to differ.
Without further ado, we’ve got the scoop on the juicy details:
- Cisco California blade servers come in a new chassis (unlike for example Intel, whose now defunct Enterprise Blade Server line consisted of rebranded IBM kit).
- The chassis has 8 slots, that can fit 8 half-width or 4 full-width blades. At the bottom of the chassis there are 4 power modules, on both sides there are two “FEX”es, or Fabric Extenders.
- Each of the 8 slots has got a 10 Gb connection to one of the two Fabric Extenders.
- The blades come in two models, both with two sockets, populated with Intel Xeon CPU’s based on the Core i7 (“Nehalem“) microarchitecture. The half-width blades have got 12 memory slots, the full-width have a whopping 48 memory banks. By leveraging the next generation Intel architecture, this allows for an unmatched memory density, which has traditionally been the bottleneck in virtualized environments. (Providing 384 Gb requires crazy expensive 8 Gb dimms. Using cheaper 4 Gb dimms still allows for an impressive 192 Gb of memory per blade.)
- There are no dedicated management or switching modules in the chassis. The FEXes connect to the UCS 6100 (UCS = Unified Computing Switch), which is based on the Nexus 5020 switch. The UCS6100 is a “top of rack” switch that will carry the TCP/IP data, block level data (through FCoE) and the management of the system. By externalizing this management and switching, Cisco makes it easier to upgrade their chassis to new technologies later on, and it keeps as much of the environment as possible close to the network core, their traditional turf, treating the server as a commodity. The UCS communicates with a management chip on each blade to manage and monitor the server components. This management solution was co-developed with BMC.
Cisco will undoubtedly emphasise how this is a solution, based on a network architecture. As Christopher Hoff puts it at his Rational Security blog: It walks like a duck, quacks like a duck, but it’s a solution.
Now, how will the market react to this announcement is unknown. HP, IBM and the likes pretend to be unimpressed by their new competitor, but it’s worth noting Cisco might be on to something:
- The more customers use virtualization, the more they need central management whereas the underlying servers become interchangeable building blocks. By integrating the management in the network switching, Cisco can reinforce their stronghold in the heart of the datacenter.
- Cisco is a new player. They arent’ experienced in the server market, but they also could start designing their solution from the ground up, learning from mistakes made by their competitors. Expect them to try to silence potential criticts with a global service organisation (in partnership with Accenture and local partners), reference customers (Savvis is named to be a beta customer) and certifications by the likes of VMware, Oracle and SAP.
- Fiber Channel and Ethernet are on a path to convergence. Storage vendors are preparing or shipping FCoE-capable arrays. (It’s likely EMC and/or NetApp will show up at the announcement later today.) Brocade, Cisco’s biggest competitor in the SAN market, knows this as well, but their aqcuisition of Foundry Networks was delayed by some shareholders. Their product lines will likely stay separate for several quarters to come, whilst Cisco is shipping a unified product line today.
- Cisco is alienating their current partners like IBM and HP. This was inevitable however, as HP was already moving into Cisco’s market with their ProCurve switches, IBM has always courted multiple brides, and is now intensifying its relationship with Juniper. In a down economy, it’s everyone to himself. It is also the opportunity for challengers to enter new markets. Cisco has got a $30 billion piggy bank, a result of their average 65% margins. They can afford a fight, and we’re likely to see one, no matter how hard IBM, HP and even Cisco want to downplay the significance of this announcement.
It’s obvious VMware is involved in this play as well. They’ve been talking about the “software mainframe” for quite a while now, Chambers introduced “Datacenter 3.0” at VMworld in 2007, and Cisco is a minority shareholder in VMware.
Update: VMware has already put out its press release and a YouTube video in which CEO Paul Maritz talks about Unified computing.
Update: The Cisco press release is out as well.
Pictures are over at Flickr, and in the data sheet.
Cisco has published a full list of partners:
Stay tuned for the Cisco announcement later today. The webcast starts at 10:30 AM Pacific Time, 18:30 CET. You can tune in at Cisco.com.
The technical figures seems impressive but no performance data was given.
It is also not clear why an IBMHP blase server would not have a similar SPEC ?
They all have nice IO Architecture and still most people prefer not to go into the excellent Infiniband type of solutions.
It looks like a big mainframe, but I’m not sure it is a software mainframe.
Seems more like a dedicated hardware solution, which may not be the best path as an industry.
Virtualization has freed us from the underlying hardware and clouds can set free even further.
This server seems like a step in the other direction, but maybe I’m missing something.
I have a sneaking suspicion that the cisco blade server is manufactured by HP. The design and the console connector looks very very similar to HP’s solution.
As the Manager of Network Systems Networking for Cisco I just wanted to add to the point made that “The more customers use virtualization, the more they need central management whereas the underlying servers become interchangeable building blocks.” Our strategy is to bring transparency to virtualization. When data center resources are connected over an intelligent, unified network fabric, things that were previously invisible become visible. As customers scale virtualization, they begin to see its inherent complexities. Policies, Management, Security, Processes, Platform ‘islands’ still exist for virtual machines just as they did for physical machines, except now the level of scale has increased by another order of magnitude. In today’s data center infrastructure, compute, virtualization, network — and even communications — resources are separate ‘islands”, and each island has a different perspective on virtualization. Cisco’s unique value is at the intersection; only the network can bring transparency and visibility across the data center ‘islands’ to enable automation and dynamic provisioning across the data center infrastructure.