C
Chris
Could someone please let me know if this is a fair assessment of NLB -
especially the last paragraph describing limitation. time for
application server?
--------------------------
Microsoft's Network Load Balancing in unicast mode works by altering
the MAC address of the network adapter hosting the Virtual IP Address
(VIP). Each host in a load-balanced cluster will have one network
adapter with Load Balancing enabled. The VIP is manually added to the
list of IP addresses for this network adapter, and each load-balanced
adapter configured with this VIP will share an identical MAC address.
When the host servers are plugged into a shared hub, they each receive
all incoming requests (usually web requests). If the incoming request
is for an allowed port on the VIP, Microsoft's Network Load Balancing
(NLB) determines which server will accept the request. If the
incoming request is for a private IP address configured on the
load-balanced network adapter (in addition to the VIP), the server
which owns that address will accept the request. As usual on a hub,
all incoming requests for an unfamiliar IP or disallowed port are
dropped.
NLB is designed to load balance any IP traffic - not just web
requests. Because it is not made with any particular network service
in mind, there are no built-in capabilities to remove a server from
the cluster if a particular service (such as IIS) is malfunctioning.
Regardless, the end-user's ability to use a particular service is the
most important thing to monitor, and these types of tests are more
valuable if they come from a remote client machine.
Microsoft released HTTPMon as a tool to monitor web sites from a
client machine outside a load-balanced cluster. When a site becomes
unresponsive, HTTPMon will send its host a WLBS STOP command. When
the web site begins responding, HTTPMon will send a WLBS START command
to its host. Since the VIP is shared between the cluster hosts,
HTTPMon must use one of the unique private IPs configured on each of
the load-balanced network adapters - otherwise, it would not know
which host was unresponsive or which to send a WLBS command.
The limitation with this system comes from the fact that IIS 5.0 will
only allow either a 1-to-1 or a 1-to-all "bind" between web sites and
IP addresses. Since the VIP must be bound to the web site you want to
load balance, it is not possible to also bind the unique private IP
that HTTPMon must use to monitor the web site without also binding all
other IP addresses owned by the server. The result is that is is only
possible to monitor one web site per cluster host.
especially the last paragraph describing limitation. time for
application server?
--------------------------
Microsoft's Network Load Balancing in unicast mode works by altering
the MAC address of the network adapter hosting the Virtual IP Address
(VIP). Each host in a load-balanced cluster will have one network
adapter with Load Balancing enabled. The VIP is manually added to the
list of IP addresses for this network adapter, and each load-balanced
adapter configured with this VIP will share an identical MAC address.
When the host servers are plugged into a shared hub, they each receive
all incoming requests (usually web requests). If the incoming request
is for an allowed port on the VIP, Microsoft's Network Load Balancing
(NLB) determines which server will accept the request. If the
incoming request is for a private IP address configured on the
load-balanced network adapter (in addition to the VIP), the server
which owns that address will accept the request. As usual on a hub,
all incoming requests for an unfamiliar IP or disallowed port are
dropped.
NLB is designed to load balance any IP traffic - not just web
requests. Because it is not made with any particular network service
in mind, there are no built-in capabilities to remove a server from
the cluster if a particular service (such as IIS) is malfunctioning.
Regardless, the end-user's ability to use a particular service is the
most important thing to monitor, and these types of tests are more
valuable if they come from a remote client machine.
Microsoft released HTTPMon as a tool to monitor web sites from a
client machine outside a load-balanced cluster. When a site becomes
unresponsive, HTTPMon will send its host a WLBS STOP command. When
the web site begins responding, HTTPMon will send a WLBS START command
to its host. Since the VIP is shared between the cluster hosts,
HTTPMon must use one of the unique private IPs configured on each of
the load-balanced network adapters - otherwise, it would not know
which host was unresponsive or which to send a WLBS command.
The limitation with this system comes from the fact that IIS 5.0 will
only allow either a 1-to-1 or a 1-to-all "bind" between web sites and
IP addresses. Since the VIP must be bound to the web site you want to
load balance, it is not possible to also bind the unique private IP
that HTTPMon must use to monitor the web site without also binding all
other IP addresses owned by the server. The result is that is is only
possible to monitor one web site per cluster host.