Optimizing-QoS-vSphere_final



Windows* VMs
It is important to understand the different
VM drivers used in vSphere so the
bandwidth shown in the monitoring tools
can be associated with the correct types
oftrafcandnetworkadministratorscan
verify that the correct one is installed.
There are three types of virtual network
adapters available for VMs in VMware
vSphere:
•vmxnet is a paravirtualized device that
works only if VMware Tools is installed
within the guest OS. This adapter is
optimized for virtual environments and
designed for high performance.
•vlanceemulatestheAMDLance*
PCNet32Ethernetadapter.Itis
compatible with most 32-bit guest OSs
and can be used without VMware Tools.
• emulates the Intel® Gigabit
Ethernet adapters and is used in either
64-bit or 32-bit VMs. It can be used
without VMware Tools.
Two other virtual adapters are available
through VMware technology. Vswif is a
paravirtualized device similar to vmxnet
that the VMware ESX service console
uses.  is a device in the VMkernel
thattheTCP/IPstackusestoserve
Network File System (NFS) and software
iSCSI clients.
On Intel® architecture-based servers
that use Intel® Ethernet controllers
and adapters, check to ensure that the
VMXNET3 Virtual Network Device is
enabled in each VM. This practice provides
the latest performance enhancements
to minimize the overhead of network
virtualization. VMXNET3 achieves a higher
throughput than enhanced VMXNET2
for a majority of the tests on Microsoft
Windows Server* 2008. Refer to the
VMware paper “PerformanceEvaluation
of VMXNET3 Virtual Network Device”
8
for more details.



Balance Loads
TheintroductionofDynamicLogical
Segmentation in vSphere 4.1 results in
subtle but important changes to best
practice 4 in the paper, “Simplify VMware
vSphere* 4 Networking with Intel®
Ethernet 10 Gigabit Server Adapters.”
1
An issue with most port-teaming solutions
isthatVMtrafcisallocatedtoaspecic
port and more or less stays on that port.
This can cause some ports to have heavy
trafcwhileothersareunderutilized.
Load-basedteaming(LBT)isanewtrafc-
management feature of the vNetwork
Distributed Switch (vDS) introduced
withvSphere4.1.LBTavoidsnetwork
congestionontheESX/ESXihostuplinks
caused by imbalances in the mapping
oftrafctothoseuplinks.Thisfeature
enables customers to optimally use and
balance network loads over the available
physicaluplinksattachedtoeachESX/ESXi
host.LBThelpsavoidsituationswhereone
link may be congested while others are
relatively underused.
LBTdynamicallyadjuststhemapping
of virtual ports to physical NICs to best
balance the network load entering or
leavingtheESX/ESXi4.1host.WhenLBT
detects an ingress- or egress-congestion
conditiononanuplink,signiedbyamean
utilizationof75percentormoreover
a 30-second period, it will attempt to
move one or more of the virtual ports to
less-usedlinkswithintheteam.LBTisan
additional load-balancing policy available
within the teaming and failover of a
dvPortGrouponavDS.LBTappearsasthe
option ,
asshowninFigure5.LBTisnotavailable
on the vNetwork Standard Switch (vSS).
By ensuring optimal utilization of all
server adapters, the load-based teaming
capability of VMware vSphere 4.1
prevents one port from being overloaded
while others may be underutilized, as
illustrated in Figure 6. Therefore, a smaller
number of ports can support a larger
amountofnetworktrafc,atthesame
time helping to ensure high levels of
network performance for the virtualized
environment as a whole.
Load-basedteamingisoneofseveralload-balancingpolicyoptionsavailablewithinthe
teamingandfailoverofadvPortGrouponavNetworkDistributedSwitch.
6
