Monday, 15 July 2013

Data center

Scalability: The Server Evolution

Scale Up    (The 1990s)                                                                                
Monolithic servers:
§  Large number of CPUs
§  Proprietary platforms
§  Proprietary Operating System
§  Many apps per server
§  High cost
§  Large failure domain


Scale Out    (Early 2000s)                                                    

Commodity servers:
§  Few CPUs
§  X86 platform
§  Commoditized Operating System
§  One application per server
§  Under-utilized servers
§  Power and cooling issues


Scale In (Now)

Blade servers:
§  Multi-core CPUs
§  X86 platform
§  Commoditized Operating System
§  Virtualization
§  Blade servers:
§  Multi-core CPUs
§  X86 platform
§  Commoditized Operating System
§  Virtualization


The first servers were designed as large, monolithic enclosures. At that time, the components were very expensive, so a single large system could service a large number of users more cost effectively than many smaller servers.
The next change in servers came with commoditized servers. These were smaller servers that cost less than the monolithic server, but they also had fewer processing resources, typically limiting them to a single service or application per physical server.  In addition, reliability concerns further drove a one-to-one server to application architecture.
Today, the push for servers is to reduce costs by increasing efficiency. Servers are being designed into smaller form factors and utilizing shared resources. Additionally, with the introduction of virtualization software, many services are being consolidated onto single physical computing resources. While this aids in many of the challenges faced by the data center, other challenges are created.



Server Evolution—Scalability

1990s
Scale Up
          
 
  Monolithic servers:
§  Large number of CPUs
§  Proprietary platforms
§  Proprietary operating system
§  Many applications per server
§  High cost
§  Large failure domain


The server scalability model through the 1990s. At that time, compute resources were typically scaled by adding more compute power to a small number of servers. The result was higher total cost of ownership (TCO) and large failure domains.




Server Evolution—Scalability (Cont.)

Early 2000s
Scale Out

Commodity servers:
§  Few CPUs
§  X86 platform
§  Commoditized operating system
§  1 App / server
§  Under-utilized servers
§  High power and cooling costs


The server scalability model during the early 2000s.  During this time, the scalability model shifted away from fewer, larger servers to more numerous, smaller servers.
This transition was primarily driven by three factors:
          A shift from centralized IT control of assets to decentralized, departmental control
          The development of clustering technologies that allowed applications to scale more efficiently across multiple servers
          Increasing standardization on the Intel x86 platform to reduce cost of ownership





Server Evolution—Scalability (Cont.)


Now
Scale In



Blade servers:
§  Multi-core CPUs
§  X86 platform
§  Commoditized operating system
§  Virtualization
§  Management complexity
§  Limited scale


Since the early 2000s, IT has begun to move away from traditional tower or rack-mounted servers in favor of blade servers. Larger servers required more space and needed to have their own power supply, networking, and SAN cables. For smaller data centers, these infrastructure requirements are not problematic, but in a large data center these requirements result in increasing complexity and cost of ownership.
By putting the computing resources of a server in a blade form factor within an enclosure, administrators gain the following advantages:
          Maximum use of physical space
          Shared power distribution
          Shared networking and storage access
          More efficient power and cooling




No comments:

Post a Comment