SN0430983-00 Rev. E 2/17 1
White Paper
Enhancing Scalability Through NIC
Switch Independent Partitioning
QLogic 57800S, 57810S, and 57840S Ethernet Adapters
Unleash the Power of Data Center Servers
Network interface card (NIC) Switch Independent
Partitioning can simplify end-to-end networking
by dividing a network controller port into as many
as four partitions, enabling dynamic allocation of
bandwidth as needed while reducing the total cost
of ownership.
SPLITTING THE 10GbE PIPE
Partitioning begins by conguring a single 10GbE port to represent as
many as four separate partitions or physical functions. Each partition is an
actual PCI Express
®
(PCIe
®
) function that appears to the system ROM, OS,
or virtualization OS as a discrete physical NIC with its own driver software,
and each partition behaves as an independent NIC port (see Figure 1).
A partition allows complete logical isolation and division of networking
and storage resources, and administrators can dynamically assign these
resources on a per-partition basis. By provisioning each of these physical
functions with the desired bandwidth in 100Mbps units, administrators can
achieve high granularity in partitioning. In essence, the technology allows
administrators to split the 10GbE pipe into several network interfaces in
whatever increments are appropriate to provide the correct bandwidth
for each task. Switch Independent Partitioning works with a variety of
standard Ethernet switches and pass-through devices.
Each partition is congured with Relative Bandwidth Weight and with
Maximum Bandwidth. The Maximum Bandwidth is used to limit the
bandwidth given to a partition, even if there is available bandwidth that
no other partition uses. The relative weight is used to determine the
relative bandwidth a partition will get with regard to the other partitions
on the same port when they compete on the same available bandwidth.
A partition will get its relative weight out of the available bandwidth but
never more than its congured Maximum Bandwidth. If a partition is not
using its congured weight, its bandwidth will be available for the use of
other partitions.
OVERVIEW
A burgeoning number of applications, rich multimedia content,
increasing virtualization, and the growth of cloud computing hamper
networking efciency. To handle these increased workloads,
many organizations have transitioned from Gigabit Ethernet (GbE)
networks to 10-Gigabit Ethernet (10GbE) networks. This bandwidth
increase can provide signicant performance and cost efciencies
while supporting efforts to consolidate data centers. However, as
organizations increasingly transition from GbE to 10GbE networks,
trafc ow for application workloads may not require the entire
bandwidth that is available in a 10GbE pipe.
Many organizations have built their IT infrastructures by adding
separate servers, networks, and storage capacity for each application
workload. These silo application deployments helped simplify
management in some ways, but left excessive pockets of unused
server capacity that contributed to network sprawl. While network
and storage convergence as well as server virtualization have helped
address these problems, the challenge for 10GbE networks has been
provisioning the appropriate bandwidth and data center resources
to meet varying application workload demands in both physical and
virtual environments. One powerful approach to help eliminate these
problems is Switch Independent Partitioning, in both physical and
virtual environments. This approach enables administrators to split
up the pipe to divide and reallocate bandwidth and resources as
needed. In collaboration with QLogic
®
, Dell
®
provides 10GbE Switch
Independent Partitioning in Dell PowerEdge™ rack, tower, and blade
servers, giving IT organizations enhanced exibility and capacity as
they migrate to 10GbE networks.
SN0430983-00 Rev. E 2/17 2
White Paper
Enhancing Scalability Through Network Interface Card
Switch Independent Partitioning
DYNAMIC BANDWIDTH PROVISIONING
Traditionally, systems administrators would oversubscribe bandwidth for
shared connections to handle demand surges by the same applications.
Instead, administrators can take advantage of the Switch Independent
Partitioning capability to handle such demands and can congure the
weighting of each partition to provide increased bandwidth presence
when an application requires it (see Figure 2). Each virtual machine (VM)
assigned to a specic partition with equal weights can acquire as much
free bandwidth as is available, while incrementally yielding bandwidth
back as demand from the other partitions increases.
SWITCH INDEPENDENT PARTITIONING CONFIGURATIONS
Each NIC partition is presented with a dedicated PCI function that
can be enabled or disabled by administrators. With this PCI function,
partitions behave as normal PCIe Ethernet devices from a host system
and OS perspective—requiring no changes to the OS. After the driver is
loaded, each enabled partition becomes essentially a dedicated network
connection or Ethernet and storage controller device. Standard device
congurations can then be applied to the partition.
Each partition can support networking functions such as TCP checksum
ofoad, large send ofoad, Transparent Packet Aggregation (TPA), multi-
queue receive-side scaling, VMware
®
NetQueue and Microsoft
®
VMQ,
along with storage Host Bus Adapter features such as iSCSI and FCoE.
Administrators can congure a partition to run iSCSI, FCoE, and
TCP/IP Ofoad Engine (TOE). All features can be enabled or disabled on
each partition as necessary.
Administrators can provision NICs quickly and easily using the Dell Unied
Server Congurator (USC).
STREAMLINING NETWORKING WITH SWITCH
INDEPENDENTPARTITIONING
Switch Independent Partitioning helps simplify the data center and the
network and storage infrastructure in several ways. For example, when
connecting servers to LANs and SANs, administrators may have to use
many cables, sometimes adding switches to reduce cable proliferation.
Switch Independent Partitioning provides a third alternative—consolidating
Ethernet and Fibre Channel connections onto signicantly reduced
numbers of devices. Like switches, Switch Independent Partitioning
reduces the number of cables without adding workloads on the network
—but Switch Independent Partitioning requires fewer devices versus
using switches and cables, and is designed to offer the following
additionalbenets:
Reduced Network Sprawl: With a two-port GbE device, a data center
can usually deploy 10 NICs, two Host Bus Adapters, and many cables
and switch ports to meet bandwidth needs. With the 10GbE Switch
Independent Partitioning approach, the same capacity and exibility can
be provided with a 10GbE Converged Network Adapter—consolidating
the infrastructure onto much less equipment while using less facility
space than when deploying GbE devices.
Maximized Network Scalability: Having a reduced number of network
devices and cables allows IT organizations and data centers to easily
scale their networks and add servers and network devices to meet
growing IT demands.
Simplied Administration: Switch Independent Partitioning can also
save time and labor by helping to simplify the management tasks for
IT administrators and enhance their self-sufciency. They can add or
replace network cards or move workloads from one partition to another
within minutes.
Figure 1. Switch Independent Partitioning Conguration