Re: Local traffic and active/standby NLB configs
- From: "Russ Kaufmann [MVP]" <russ@xxxxxxxxxxxxxxx>
- Date: Tue, 18 Apr 2006 09:41:42 -0600
"Ken L" <kenl@xxxxxxxxxxxxx> wrote in message
news:e66hOjmYGHA.3448@xxxxxxxxxxxxxxxxxxxxxxx
Thanks for responding, Russ. Regarding your questions . . .
NLB doesn't work like this, normally. You can set it up, though, using
Priority for Affinity. It basically defeats the purpose of NLB, in my
opinion, but you can do it.
I am aware that normally, the servers are configured to handle all
workloads. In this case, though, I have a pair of servers and a number of
different workloads with radically different bandwidth requirements. With
the router and dual ISP configuration I have available I can optimize and
balance the bandwidth and performance by segregating specific workloads to
specific servers. If I have a server failure, though, I want failover to
be automatic. If this happens, all the workloads will be combined on the
surviving server. Each of the primary services will be cross-configured
with its corresponding hot spare on the opposite machine. Some services
will run their primaries on one machine, the others will run theirs on the
opposite machine. Configuring NLB in single-host mode allows me to keep
the backup services, in each case, running and on hot standby. If a
server fails, I don't have to change any IP addresses to bring up the
backups.
Are you saying that you want Host-S to be a client to Host-A while it is
part of the NLB cluster?
Yes. I know it sounds bizarre, but there is a rationale to it. We run
telemetry devices on other peoples networks. To minimize configuration
hassles (firewalls, proxy servers, etc.), we use a client program that
talks the HTTP protocol to transmit the data from these remote sites. To
the security infrastructure at the sites, our app looks like a standard
web browser. Most locations allow port 80 traffic outbound without any
restrictions and it eliminates having to open specific ports through
firewalls just for our traffic). On the server side, we use PHP scripts
on regular web servers to receive and reformat the data, and then send it
to our main, home-grown services which handle the data as UDP packets
through a specific port.
What will happen when the web script sends the UDP packet to the home
grown service, which will have a primary and standby version running on
the two servers, sharing a NLB cluster address. The web servers were going
to be set up as "traditional" round robin NLB services across the two
machines. My concern was when a web transaction comes in on the standby
HOST-S server and sends the UDP transaction to the 192.168.30.10 address,
whose active server is over on HOST-A. If the network stack takes a look
at the IP address and sees that it is local (which it is because both
hosts share the cluster address) and bypasses the NLB intermediary
selection processing, it will get processed by the standby host rather
than my intended primary server. The NLB drivers get in the middle of
normal NIC-to-TCP/IP stack processing (to handle things like duplicate-IP
detection). I'm just trying to figure out whether they will also enforce
the cluster filter rules by forcing the UDP transaction out through the
NIC's so all machines in the NLB cluster will see the UDP transaction.
Thanks again for the help.
Ken
"Russ Kaufmann [MVP]" <russ@xxxxxxxxxxxxxxx> wrote in message
news:ObWRlglYGHA.4996@xxxxxxxxxxxxxxxxxxxxxxx
"Ken L" <kenl@xxxxxxxxxxxxx> wrote in message
news:uho6ybkYGHA.3444@xxxxxxxxxxxxxxxxxxxxxxx
I am getting ready to set up a pair of servers running a NLB cluster.
The NLB clusters will be run in single-host filtering mode
(active-standby config) to provide failover.
Lets say one host is named HOST-A (for active) and the other is named
HOST-S (for standby) and the NLB cluster has a cluster IP address of
192.168.30.10.
NLB doesn't work like this, normally. You can set it up, though, using
Priority for Affinity. It basically defeats the purpose of NLB, in my
opinion, but you can do it.
My question is what happens if one of first app, which happens to be
running on HOST-S, trys to connect to the cluster address,
192.168.30.10, while HOST-S is in standby mode for the second, cluster
based app?
Are you saying that you want Host-S to be a client to Host-A while it is
part of the NLB cluster?
In a non-NLB environment, I've never seen the packets show up on the
Ethereal trace because the TCP/IP stack handles the connection
internally (without involving the NIC for traffic on local adapters).
In a NLB environment, does the TCP/IP stack behave differently and force
the transaction out through the NIC so it participates in the NLB
selection process, or does it send it directly to the load-balanced app
(even though the app is not expecting the transaction because it is on
the Standby side of the N:LB pair)?
You are going to need to test this. It can depend on the number of NICs
per node, how they are configured on the network, how they perform name
resolution, and so forth.
Ken,
You might do better if you talk about what your business requirements are
and how they can best be met.
--
Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp
.
- References:
- Local traffic and active/standby NLB configs
- From: Ken L
- Re: Local traffic and active/standby NLB configs
- From: Russ Kaufmann [MVP]
- Re: Local traffic and active/standby NLB configs
- From: Ken L
- Local traffic and active/standby NLB configs
- Prev by Date: Re: Local traffic and active/standby NLB configs
- Next by Date: Re: Different of Windows 2000 cluster and Windows 2003 cluster
- Previous by thread: Re: Local traffic and active/standby NLB configs
- Next by thread: NLB web servers sharing a common file store and DFS
- Index(es):
Relevant Pages
|
