High Availability & Removing Single point of failure


From a high availability perspective, how is it possible to set up a load balanced Service Manager Application Server cluster environment, so as to be made fault tolerant and without single point of failure?

My understanding when using a Horizontal Scaling implementation with multiple application servers, that one separate server must act as a load balancer and is considered to be the only single point of failure, which does not meet the objective of a high available design (only load balancing but no failover if the load balancer server fails).

Thank you.
  • In theory can use application clustering technology e.g. Win 2003 server cluster. Have not seen it actually it done nor am I aware of a published white paper on the subject. HP claim to have done it with SC but never asked about SM (theory should be the same).

    Do you have any real life statistics to show that SC/SM load balancer is unstable or are you thinking ahead? If it only happens rarely then is it a big issue if access to the desk is down for 15-30min until a support team is called to look it to it?

  • I'm not implying that the load balancer is unstable. Rather, I read somewhere in ServiceCenter (not SM although I assume it's the same concept) that the machine or host running the load balancer process is considered as the only single point of failure in a horizontal scaling implementation.

    Example, if I have three machines, machine 1 acting as a load balancer that routes or load balances the incoming connections to the machines 2 and 3, it means I can keep running if machines 2 or 3 fail (even though I'll support less connections/threads at the time), but if my load balancer machine 1 fails then the entire system has failed because I can't route to machine 2 or 3 anymore.
  • You reading is correct as I understand. Only option at present that I understand to be viable is to use an application cluster to do a real time failover of the SC server running loadbalancer/system listener/ system processes if a problem is detected.

    I suspect it more hassle to build/test and maintain that its worth for a Service Desk which in most cases can cope with occasional downtime as long as its not happening too often. Don't have any stats to hand as we are still vertical scaled though horizontal scaled is on the near term roadmap.

    Has anyone went to the hassle of doing this with servlet implementation on SC or SM in the wild? Has anyone got evidence of instability in the wild?
  • Is there any official document by HP that explains how to perform the clustering for Service Manager (or ServiceCenter) at the OS level? Is it officially supported by HP?
  • Hi Anis,
    There is no official document as such for clustering of HP SM.
    But, We have configured HP SM on windows cluster environment. We have installed HP SM on two seperate nodes of cluster. The MS SQL 2005 is again on cluster. The HP SM will be running only one node at a time. Once the node goes down, the SM comes up in another node automatically. Hence no single point of failure of SM. It is up and running at customer place at present. If you want I will share more on this. Please let me know the same.

  • Well, how did you configure the Service Manager application server to work with Windows OS clustering? What are the changes that must be applied to the sm.cfg/sm.ini, or is the configuration entirely from the OS side only?

  • Sree, can you provide any more info, please?
  • Hi Sree

    Can tell me,what parameters to be set on both server in sm.ini for configuring windows 2003 cluster
  • Hi Sree,

    Could you share with us how SM is configured in a Windows 2003 cluster environment?