|
 |
|
General Information
|
Add To My Personal Library |
March 30, 2007
• Vol.29 Issue 13
Page(s) 28 in print issue
|
Scale-Out vs. Scale-Up
Which Technique Is Right For Your Data Center?
|
Choosing an enterprise architecture for your corporate data center includes much upfront analysis and planning before you settle on scale-out or scale-up architecture. These two popular architectures both have particular advantages and disadvantages for the small to medium-sized enterprise building out its data center.
Scale-out architecture enables a data center to grow over time. This HA (high availability) architecture uses multiple machines and infrastructure working in tandem. Typically, Ethernet provides connectivity between machines in the scale-out architecture. Key benefits of scale-out architecture include redundancy and high availability of applications.
Scale-up architecture can be more expensive because it uses multiple processors sharing the same memory space, according to Daniel Cox, manager of the High-Performance Computing Programs for the Industry Standard Server group for HP (www.hp.com). This architecture relies on denser clusters of servers such as blades to drive computing tasks.
 Choosing Between Scale-Out & Scale-Up
When considering an enterprise architecture for a new data center, other factors come into play, including power and cooling. Cox advises that you plan five years out based on projected business growth.
Rik Wright, group product manager for the Windows Server Business Group at Microsoft (www.microsoft.com), says, “Launching a new data center definitely presents the opportunity to rethink your network capacity and footprint, where the scale-out vs. scale-up decision is primarily driven by application architecture, processing, and data warehousing requirements. Today we see a data center environment where innovations such as smaller hardware form factors, lower power consumption of computing devices, and increased scalability of multicore chipsets are all definitely lending themselves to more robust and dependable scale-out architectures.”
Another factor that needs addressing is a capital investment plan. Building out a scale-up architecture requires a larger upfront investment because the servers are more expensive. Building out a scale-out architecture can be done incrementally, pushing the investment costs across multiple business quarters.
Considerations
Wright says, “The first consideration I would call out between scaling out and scaling up is how your business intends or needs to deploy particular applications. For example, do you have need to run an application on a single machine such as a multi-terabyte database, or is the application intended to be spread across several servers, such as collaborative application suites like Microsoft Exchange or other easily dispersed workloads such as Web serving?”
Wright continues, “The second consideration I would highlight is application availability. While large scale-out servers tend to be highly reliable, there are often significant maintenance costs associated with maintaining that level of uptime, and yet you still have a single point of failure for whatever service that server is providing. If the application you are deploying allows for a distributed architecture, you can cluster several smaller servers together and present them to the network as a single service, therefore providing fault tolerance if one server fails without bringing down the overall availability of the application.”
According to Wright, “The third consideration that deserves being called out is your server management infrastructure. . . . Each added server in a scale-out environment logically increases the footprint of the network and therefore possibly the number of staff to manage it. However, in today’s volume server market, we are seeing a lot of new technologies in the areas of systems management automation and virtualization capabilities, such as our upcoming Microsoft Systems Center product, that provide for easier, simpler, and consolidated management of scale-out server topologies.”
Scale-up architecture, with its reliance on more powerful processors and compact configurations such as a blade server, means that while blade servers are going to take up less space, their processors run hotter than their traditional server counterparts. This means data center cooling needs to accommodate the additional heat from the blade servers. Maintenance on blade servers can also be more expensive.
The trade-off with scale-up architecture is that while the dense architecture requires less physical floor space in your data center, cooling costs could grow.
Scale-out architecture, with its reliance on multiple traditional servers, requires more floor space, and all the while it may not be as fast as a scale-up architecture. It is easier to add new technologies incrementally. Additionally, scale-out architecture is based on pairs of machines each backing up and protecting one another.
There is also a middle ground in scale-out and scale-up architectures where you can leverage each architecture method. Cox suggests exploring placing your applications on blade servers (scale-up architecture) for performances and placing your storage in a scale-out architecture.
The main deciding factors include high-availability (check your SLAs), cooling, and space requirements. 
by Will Kelly
|
Scale-Out & Scale-Up Architectures Compared
| Scale-Out Architecture | Scale-Up Architecture | | • Uses multiple machines • Provides redundancy for mission-critical applications but with some sacrificing of performance • Capital investments can be spread out across multiple budget cycles • Easier architecture for the incremental introduction of new technologies | • Uses multiple processors sharing the same memory space • Some applications work better in a scale-up architecture environment • Larger upfront capital investment • Less flexibility for the introduction of new technologies | |
|
|
