CPU and Memory Scalability
With a single-core CPU, CPU is usually the bottleneck.
Without virtualization, multi-cores are often
underutilized.
With multi-core and virtualization, memory is often the
bottleneck.
Today, server processors are multicore, offering multiple
processing cores in the same space that was previously used for single-core
processors.
In addition, many vendors have created memory controllers
that have the ability to address large amounts of RAM into the tens of
gigabytes. While blade server sizes have remained the same, the processing and
memory has increased significantly, allowing blade servers to run processing
and memory-intensive applications.
Virtualization
Virtualization has allowed companies to more easily
consolidate servers in the data center. Instead of the standard one-to-one
server model, many servers—each running in independent “virtual machines”
(VMs), can run on a single physical server.
This is performed by creating individual virtual hardware
that can function like a standard server. There are a number of advantages to
using virtualization, including better use of computing resources, greater
server densities, and seamless server migrations:
•
Virtual
machine: A virtualized set of hardware that is able to
operate in a similar fashion to a physical server
•
Virtual
server: A virtual set of hardware along with the operating
system, applications, and files that is able to operate comparable to a
physical server
•
Hypervisor
layer: A software layer that abstracts the physical
hardware and creates individual virtual hardware for each VM. For example, VMware ESX, Microsoft Hyper-V,
Xen, etc.
Management Challenges
Workload Portability
Servers have unique identifiers that identify it on
various networks. These identifiers are
tied to the hardware. If you change any of these items, the server will potentially
lose its ability to access network resources, or even boot an operating system.
These identifiers include:
•
World wide name (WWN):
Hard coded to a Host Bus Adapter (HBA), this
identifier is needed for SAN access.
•
MAC address: Hard coded to a network interface card (NIC), this
identifier is needed for LAN access.
•
BIOS: This identifier contains settings that are specific to
the server hardware.
•
Firmware: This low-level software runs on peripheral device and
adapter cards to enable the operating system to interface with the device.
If a server fails and its operating system and
application needs to be migrated to another physical server, the operating
system/application and the network may require manual configuration
changes. These manual configuration
changes lead to longer recovery times and increased application downtime.
The alternative to manual configuration changes is the
use of “stateless servers.”
•
Stateless server: A stateless server is a server whose operating system and
application personality have no ties to the physical hardware. One way in which to accomplish this task is
to use transportable, virtual, unique identifiers.
•
Server personality: A server personality is the operating system
configuration and application settings.
This is a fully functional set of programs, files, and settings required
to perform the given task of the server.
Server Management
As computing capacity increases within the data center,
so does complexity. Blade servers solve
many issues but they also bring an additional point of management into the
datacenter per chassis. Several
independent systems must be managed including LAN, SAN, servers, and
storage. These separate resources must
be managed at each network layer—access, aggregation, and core. Typically, these resources are managed by
individual teams and may be monitored using proprietary system-monitoring tools
and alert aggregators. In many cases,
customers must use multiple monitoring applications to cover all aspects of the
data center.
Data Center Network
The data center consists of multiple networks. In
addition to the LAN and the WAN, 20 percent to 40 percent of all servers are
also connected to a SAN over Fibre Channel. The use of high performance
computing is also growing; for example, with financial trading applications
that demands very low latency. This leads to more complexity with regard to
cabling, power, cooling, and management.
The increase in server quantity has resulted in a similar
increase in network requirements. As
application demand expands and new servers are implemented, the networks must
grow to meet this demand. These networks
include LAN, SAN, and High Performance Computing (HPC). This not only results in increased power and
cooling needs but also increased cabling costs and management challenges.
Networking needs vary greatly depending on the operating
system and application but can reach as high as eight 1-GE ports and four 4-G
Fibre Channel ports for a virtualization platform. This is up to 12 cables per physical server
and can add up to 252 cables for a rack filled to capacity with 2U servers.
Blade Challenges
Blade Benefits
§ Reduction in redundant equipment costs
§ Power and cooling savings
§ Shared switching could reduce cabling
§ Rapid hardware provisioning
The consolidation of discreet servers into blade chassis
provides a number of advantages in the data center:
•
Reduced physical footprint
•
Shared networking and SAN switching
•
Reduced cabling
•
Rapid provisioning of additional resources
Blade Challenges
§ Increased physical density may create power and cooling
challenges
§ Increased compute density could increase bandwidth and
cabling requirements
§ Each chassis and local switching creates additional
management points
As the physical density of a blade environment increases
relative to rack mounted servers, there may be more power and cooling required
in the same physical space. This can
cause challenges in powering and cooling the data center.
The relatively high density of compute resources in a
blade environment often requires significantly more bandwidth and cabling to
fully utilize the deployment.
Each chassis in a blade environment will typically have a
management IP address and require monitoring for system and blade health. Additionally, each local LAN or SAN switch
(if installed) also adds management overhead and overall network design
challenges.
Summary of Data center and Data center part 2
§ The data center of today is the product of multiple
evolutions in thinking and architecture.
§ The scale-out x86 architecture is currently undergoing
another evolution through the use of blades, virtualization, and I/O
consolidation.
§ New technologies and architectures can add infrastructure
and management complexities.