I think the disconnect is coming from an understanding of what
"distributed" means to Spectrum.  This has been interpreted many ways
and is actually described well in the docs, but can be difficult to
grasp the full concept.  If possible, I highly recommend going to CA
training because you can play with the concept and work with the trainer
to try different scenarios.  I have been working with Spectrum for many
years and still have to think about the pieces.  Now that I am done with
that suggestion, I'll try to answer what you have here.

 

Whenever you have more than one SpectroSERVER that are communicating
with each other (other than as a backup), you have a "distributed"
setup.  The SpectroSERVERs don't forward alarms to each other (except in
special situations I won't go into here) it is up to the client
applications to get the alarms from the different servers.  So OneClick,
or in this case AlarmNotifier registers with the SpectroSERVERs to
receive alarms from them.  You can't setup a daisy chain.  It may appear
that way if the AlarmNotifier is running on server A1 and is processing
alarms from Server A1 and B1, however what AlarmNotifier has done is to
actually register to received alarms from both servers and process the
requests that come from each SpectroSERVER independently.

 

Now when you add fault tolerance, it gets a little more tricky, but the
concept is still the same.  With fault tolerance, A1 and A2 both have
the same landscape handle but different "precedence" (likewise B1 and
B2).  Client apps like AlarmNotifier will listen to the server with the
lower precedence (A1,B1) unless it looses contact to that server and
then will connect to the server with the higher precedence (A2,B2) until
it can contact the "primary" server again.

 

Now to try to simplify all that:

On A1 you have AlarmNotifier running.  AlarmNotifier connects to
SpectroSERVERs on A1 and B1 normally.  If AlarmNotifier can not contact
the SpectroSERVER on A1 it connects to the SpectroSERVER on A2.  If
AlarmNotifier can not contact the SpectroSERVER on B1 it connects to the
SpectroSERVER on B2.  It is completely up to AlarmNotifier to handle
which server it receives alarms from.  The alarms from the
SpectroSERVERs on B* are not processing through the SpectroSERVER on A1.


 

Now if you want to have multiple AlarmNotifiers running and feeding
NetCool, you will have to build your own logic to remove the duplicate
alarms.  One suggestion is to have some type of header that indicates
which AlarmNotifier the alarm comes from and then use Impact to sort
through all that.  

 

Hopefully this helps you pull the concepts together.

(Note for those not familiar with NetCool - the NetCool probe uses a
"captured" AlarmNotifier to gather the alarms)

 

 

A cheaper suggestion - Investigate CA's new product Event Integration
Service as a replacement to NetCool.  We are in the process of replacing
NetCool with EIS.  We were stuck with NetCool due to the need to enrich
alarms with data from an external database, but now with EIS we can do
that an more.

 

Bill Barnes

 

________________________________

From: Calvin Lane [mailto:[email protected]] 
Sent: Tuesday, April 07, 2009 5:58 PM
To: spectrum
Cc: spectrum
Subject: Re: [spectrum] Is this failover scenario feasible?

 

 

Thanks for the great response Cristi.  The 2 SSs will not be configured
to be distributed.  They will actually be seperate pollers with just one
SS fowarding alarms to the other SS (at least this is what the customer
is looking for).  As a follow up question, can SSs be 'daisychained' in
that fashion?  What are the various methods for sending alarms between
SSs?  Also, true the Alarm Notifier is not fault tolerant?  But, if you
have the same policy/filter configurations on both the primary and
secondary, then they should both behave the same when either box is
active.  Correct? Thanks again.

 

Calvin

On Tue, Apr 7, 2009 at 5:02 PM, Mitrana Cristian
<[email protected]> wrote:

 Calvin Lane wrote:

Hello everyone,

I have a customer that wants to design a failover/fault tolerant
soluction such that all SPECTRUM data/events/alarms are funnelled
through
one primary SpectroServer that forwards alarms over to a Netcool object
server.  Please refer to the attached diagram and let me know if this
conceptual design is feasible.  Thanks.

 

Using Spectrum just to send alarms kind of defeats the purpose ... as
its primary function is to act as a NFM and not as a NPM.
 The above scenario implies 2 SS with failover backups (distributed ?).
As it was discussed in the past on the list, the Alarm Notifier (which I
assume you must be using it to forward alarms to Netcool) is not fault
tolerant so you really don't have an option in the proposed scenario.
 Maybe if you monitor the failover process (from Netcool ?) you could
trigger Alarm Notifier start on the failover and then stop it on
failback. Look under the 9.0 documents for failover to see how this kind
of monitoring could be done.

hth,
-- 
Cristi Mitrana

---
To unsubscribe from spectrum, send email to [email protected] with the
body: unsubscribe spectrum [email protected]

 

*       --To unsubscribe from spectrum, send email to [email protected]
with the body: unsubscribe spectrum [email protected] 

 


---
To unsubscribe from spectrum, send email to [email protected] with the body: 
unsubscribe spectrum [email protected]

Reply via email to