For a long time x86 servers have been a commodity…something no one really paid attention to and innovations were few and far between. They have gotten smaller in size…from 6U or 8U to mostly 2U and 1U…but a server today is pretty much what it was years ago with the old beige Proliants. The only real difference was the introduction of blades. Traditional blades can be good and bad, depending on your viewpoint. But when you really get down to it most blade enclosures are just mini-racks. They have servers, power, storage fabric, and network switch fabric. You took the large racks and put them in a chassis to allow for high-speed backplane communication and, supposedly, easier management. Even so…they are still pretty much consolidated racks.
I think we’re starting to see a real change in mindset when it comes to server form factor and connectivity. The first look at this is Cisco’s UCS blade system. Cisco resolved a number of problems that many people had with traditional blades. They made the chassis very cheap. They removed the switches from the chassis. They consolidated management across different chassis. They integrated true consolidated fabrics for storage and data. In essence, they’ve done what a lot of us have been after for a long time….
A good and long time friend of mine, Jacob Hall, has been discussing his concept of a Processor Array for many years. The fundamental idea was to separate everything from a server that tied it to a specific task or identity. Give the physical server the ability to be anything you need it to be at any one time. Create a large resource of compute power just like we have now with large resources of storage on SANs. Maybe you need a HPC grid at night but performing other tasks, such as running VMware ESX, during the day when load increases. This has been reasonably possible for several years by fitting together a large collection of tools and custom code, but not something that many people want to attempt…until now. With the combination of Cisco UCS and it’s included features such as service proviles plus tools like EMC’s Unified Infrastructure Manager (UIM) we are very close to having this today using off-the-shelf components and software. You will be able to very easily change the personality of your servers depending on the need at the time. UIM will deploy the storage and configure the environment (such as zoning storage fabric) on-the-fly. Isn’t this the private cloud that we’ve all been wanting and working toward?
One problem that Jake and I had with deploying blades in large environments was the cost of the chassis. They were too expensive to factor in as Layer 1 expenses along with racks, power, and cabling. Instead it was up to the applications group to purchase their own enclosure and then no one wants to share or be stuck buying that first blade. The Cisco UCS model resolves this. The chassis is cheap enough to pre-deploy. Groups can buy blades and have them installed without needing to buy other infrastructure. No one feels cheated.
Will Cisco be the only one? Absolutely not. HP is gearing up for a real fight in the server and consolidated fabric market. When HP bought 3COM many people scratched their head and wondered why…and it quickly fell off the news radar, but just wait. The reason Cisco could do such a great job delivering UCS was the fact that the new secret sauce was really in the network and management side, something Cisco has been doing for a long time. HP, while having experience in that arena, hasn’t been known as an innovator. Now with the combination of HP and 3COM I’m very interested to see what they do. I just envision HP and 3COM engineers working around the clock on their next generation consolidated blade offering. The only thing bothering me right now is HP’s attempt at FUD, but maybe this just shows how competitive this market is and will be.
It’s fun to be in to server infrastructure again. I love disruptive technologies and that’s what I think we’re seeing here. While this type of dynamic infrastructure isn’t needed by all, it will help many organizations reduce the size of their environments while scaling and evolving much faster than before.