Our recent blades vs. rack servers face-off between Chris House and Rick Vanover has spurred some hot debate. Virtualization blogger Aaron Delp, a senior engineer at ePlus Technology, submitted this response on the topic of blades vs. rack servers:
Here is how I see the rack vs. blades debate: No solution is right all the time! There are situations in which racks are better, and there are situations where blades are better.
- Everyone’s prices are different based on a million different factors. No one pays list. Cost comparisons must be made on a customer-by-customer, situation-by-situation basis. That being said, from my experience, blades will be cheaper overall once you meet a break-even point — usually in the five- or six-blade range.
- Power can be difficult to understand, but the following statement always holds true: The more blades you have in the chassis, the more power-efficient it will be. Blade chassis power supplies are designed to be at least 90% efficient; most rack servers are 75% at best. (See part one and part two of my articles on the calculations involved.)
- Management of BIOS and firmwares: With the exception of the Cisco UCS platform, both racks and blades are still flashed in pretty much the same way. There isn’t a huge advantage to either platform. Cisco UCS and its stateless model really changes the game and should serve as a model to all other server vendors.
Racks are better under the following conditions:
- Small number of servers: If you are only buying a small number of servers (three or four, maybe), the savings brought by blades are negligible.
- Power constraints: Some older data centers and many co-los don’t like or can’t support 208V power.
- High I/O requirements: I see this going away as 10G and converged fabrics come into play, but sometimes you need so many I/O cards that a blade solution is not appropriate.
Blades are better under the following conditions:
- Consolidation of the chassis and one-time setup: Install the chassis in the rack, and you have consolidated cable runs for KVM, power, networking and FC. This reduces the costs of wiring everything up every time a server is added and saves time.
- I/O Virtualization: All major blade systems (HP, IBM, Cisco) can now virtualize the MAC address and FC WWPNs. This leads to a true “wire-it-once” solution. There’s no need to involve the SAN and network teams to redo VLANs or LUN masking because of a blade hardware change.
- Power efficiency: As I covered above, blades will always be more power efficient above a certain number of blades.