QLogic 8200 Series The Value of Full Hardware Offload in a Converged Ethernet Environment - Manual (Page 3)

hSG-WP10004 SN0330923-00 rev. B 09/11 3
White PaPer
although virtualization is driving consolidation, it also imposes additional
overhead on the CPU, network, and i/O. Virtualized environments use the
hypervisor to perform software emulation to abstract it resources away
from the physical hardware. this method comes at a considerable cost,
as CPU overhead is incurred because virtual resources need to be mapped
to physical resources. additional overhead in a VM scaling environment
further compounds the problem. For example, consider a single physical
processor core within a server that supports multiple virtual processors.
When there are 12 or more VMs installed per physical processor core,
a single physical processor is now hyper-threading across many virtual
machines, hindering the application within the virtual servers from scaling
linearly.
With limited CPU resources, it is not logical to further burden the CPU with
processing i/O storage requests, which can be done with the addition of
an offload adapter, such as the QLogic 8200 Series Converged Network
adapter. to put this into perspective, consider a video card in a personal
computer used for gaming that offloads the video processing from the CPU.
Gamers are well aware of the advantages of offloading the video processing
and using graphic adapters to conserve all available CPU resources for
the video game itself. advantages of the video card include stutter-free
video playback, vibrant high-definition images, and, most importantly, CPU
acceleration to support the most performance-hungry gaming applications.
When it comes to an enterprise-class application, an Open FCoe driver
solution not only fails to meet enterprise requirements but it also can be
detrimental to data center virtualization goals, a targeted environment for
FCoe. Data center consolidation through virtualization will require CPU
horsepower to efficiently scale VMs and provide the i/O bandwidth needed
for enterprise applications. this can only be provided by an adapter with
offload capabilities.
Enterprise Requirements for FCoE
Converged
Network
Adapter
Software
Initiator
Concurrent i/O Support for Consolidation Yes No
Scalability within Virtual Operating
environments
Yes No
Support for i/O intense applications Yes No
Data integrity assurance Yes No
efficient CPU Utilization Yes No
iOPs Scalability Yes No
enterprise reliability Yes No
investment Protection Yes Yes
Broad OS Support Yes No
interoperability Yes No
The Underlining Cost of Open FCoE
the first thing you may hear about software initiators and the Open FCoe
initiative is that it’s free. While there may be some truth to the initial savings
from a NiC compared to a Converged Network adapter, it’s important to
consider the hidden costs of a software initiator. For example, a software
initiator will consume approximately 1Ghz of processing power per instance
managing FCoe transmissions for full duplex read/write operations. With
today’s multi-core processors this may not seem like a problem, but
combined with server virtualization, where a single processor core is being
shared by multiple VMs, each VM will require 1Ghz each just to process
storage and data networking requests. in enterprise environments, this
leaves inadequate CPU capacity devoted to virtual servers and business-
critical applications that run on the same server. QLogic 8200 Series
Converged Network adapters free these valuable CPU cycles, improving
the computing system’s performance without the cost of upgrading or
adding additional CPU(s). this will allow you to run more applications per
server and extend the useful life of your servers. additional cost factors
should be considered as well, and have been summarized in the following
information.
Operational efficiencies
the aforementioned software initiator consumption of 1Ghz of an
embedded processor to process FCoe transmission for full duplex read/
write operations can be reduced up to 88 percent by installing a QLogic 8200
Series Converged Network adapter with offload capabilities. the adapter
can increase operational efficiency because more VMs can be supported
per CPU; an increase in throughput is delivered per adapter, which allows
increased scalability for applications and VMs. the same QLogic adapter
can be used to offload iSCSi and provide ethernet connectivity, replacing
three adapters with one, further increasing operational efficiencies.
CPU efficiency
Moving to a converged infrastructure can provide significant advantages,
but relying on server resources to drive a protocol processing request
is a less than optimal solution. if users believe that some applications
will require a higher i/O load, and if they want to share the infrastructure
with the rest of the servers and applications, moving to a converged
infrastructure can provide these advantages. however, through Open FCoe,
software initiators will use server CPU resources to complete processing
of all supported i/O protocol stacks. Many claim that with the low cost of
processors today a valid option would be to throw cheap CPU resources
at i/O requests. this does not necessarily make sense; it shops need to
evaluate which applications require the advantages of FCoe networks, and
which operating systems and applications will be running on them. When
doing this, server virtualization appears to come out on top.