I’ve been watching enterprise IT for over 20 years now, and I’ve seen some radical changes. In the server space, one of the biggest shifts was the form factor of the servers: From tower to rack-mount to blades. But what makes a blade server anyway? Let’s consider this for a moment, as we watch another shift in progress.
Blade servers are easily recognized in the data centers, trade shows, and product catalogs of today: They’re the ones that nestle together in an enclosure, sharing some resources rather than standing on their own in a rack or on the floor. But what is the essential element that separates a blade from any other kind of server?
Let’s consider some defining characteristics of the blade server species:
- Physically, a rack-mounted blade enclosure contains a number of server blades. Each blade is smaller than a 19″ rack, allowing more to be packed into the same space. This contrasts with larger, self-enclosed tower or rack-mount servers.
- Server blades rely on the blade enclosure for critical supporting functions like power, cooling, and I/O ports, so they cannot stand alone. This further improves density and efficiency. Stand-alone servers, on the other hand, include these functions in their case.
- Blade systems include some sort of management device that monitors the blades and can control some of their functions. Conventional servers do not always have this kind of consolidated management.
- The blade chassis includes consolidated and shared I/O channels, ranging from keyboard, video, and mouse (KVM) to networking and storage (usually Ethernet and Fibre Channel). These add flexibility, since external ports can be shared by multiple blades and reconfigured without disruption.
- Blade systems are optimized for high availability, with hot-swap components everywhere from power to fans to the blades themselves. Since these are shared, it is more efficient to purchase redundant parts for a blade system than for each server individually.
To me, these five elements are key to a modern blade system. Without them, a blade solution cannot meet the expectations of buyers (or the promises of vendors!) And what are these benefits? I took a look at the marketing materials for the leading companies in the space (HP, Dell, IBM, and Cisco), and this is what they promise:
- Efficiency – More processing power in a smaller footprint (physical size, power consumption, cooling, and weight)
- Manageability – Simpler and cheaper systems management
- Reliability – High availability thanks to redundant components
- Performance – Improved I/O performance thanks to shared network and storage features
- Flexibility – Simplified cabling that can be reconfigured in software
Ooh, I can’t wait for more! Good post.
The shared management and consolidated I/O requirements disqualify some multi-node platforms, like Dell’s C5000, IBM’s iDataPlex, SuperMicro’s MicroCloud, HP’s SL series, and others. Are you intentionally categorizing those as non-blades — perhaps the same genus but a different species?
You covered a lot already
This might fall under Manageability but all offer a
form of
virtualization abstraction . The ability to
replace parts or move hardware within the ecosystem without reconfiguration.
Also most blade systems can control power on sequence, something you can’t do with a whole rack of stand alone 1u servers if the UPS fails
That’s all for now.
That’s all for now.
I’m not trying to enforce a definition so much as get my head around what a blade server is and isn’t. Yes, there’s a new generation of hyper-scale servers that are a separate species. That’s article 3 in this series! Are they blades? Or something else?
Great write up – and a question that is going to be more difficult to answer in the future. With more companies offering multi-node servers that share power, cooling and even management, the lines between rack servers and traditional blade servers are blurring. “What is a blade server”? I’m sticking with the traditional concept of blade server: a server that shares infrastructure including power, cooling, management and I/O connectivity. You’ll see that the biggest difference between the traditional blade server and the multi-node is in I/O. Blade servers share I/O whereas multi-node systems do not. Multi-node systems are created for cloud and HPC therefore they don’t usually require a lot of NICs and HBAs so providing a shared I/O platform technically doesn’t make sense. So if they aren’t sharing the I/O – they aren’t considered blade servers (in my personal opinion.)
You may have seen where I’ve written what the blade server of the future would look like, but in summary I think everything will be modular – CPUs, memory, storage, I/O – all will be their own node or module. Maybe it’s a pipe dream, but it’s where the industry is headed.
It’s interesting to contrast a “definite blade” like the HP C3000 with a “shared server” like the Dell C5000. There was a great comparison of these systems somewhere, but now I can’t find it!
Anyway, the key difference seems to be management, I/O consolidation (as you point out), and high-availability features. Blades have all this, hyper-scale servers don’t.
Thanks for the comment!
You guys consider Nutanix a blade server? I think it fits even though it’s not as tradtional as the major vendors.
I do not consider @Nutanix to be a blade server. It’s a different animal entirely! I’m actually working on the Hyperscale Server article in this series right now!
Great article – I’ve never been quite sure on what a Blade server actually is, as I don’t work in that part of IT, but this article clears it up. Thanks!
Tried and trusted tower servers have been providing server reliability for years. If you’re looking for the latest in upright, stand-alone units, IBM express range of tower servers deliver exceptional value, performance and ease of use.
I find it interesting to price out a blade server from newegg with solid state drives. Prices are dropping quickly!