Providing redundancy for processes on different servers

Question

Say I have for example 5 physical machines with similar resources (CPU speed, RAM and storage) named s1, s2, s3, s4 and s5.

If I have a processes running on s1 but then s1 fails somehow, how can I automatically run that process on one of the other 4 servers? Would that require shared storage across all these servers so the process can essentially just be started again on another machine?

Quick question slightly related also. If I have 5 IP addresses available for these 5 machines can I essentially assign those 5 just to the block of servers so each one can access each IP individually? So if a process runs and is accessed on xxx.xxx.xxx.1 but then s1 dies it swaps over and runs on s4, could s4 then have access to that IP address before so a user only sees a brief interruption in their service?

Answer 1

It is going to depend greatly on the application. Just the IP switch is easily done through linux virtual server (http://www.linuxvirtualserver.org/)

You assign it a virtual IP and it redirects the connection automatically. You can either just have it make a tcp conenction, if the server is up, it is considered active, or you can have it do a evaluation on a response. Ours checks the content of a html request to evaluate if the server is up or not. Failover is seamless for most cases.

This is perfect if the target app is say MySQL in master-master mode or a web server where the actual storage is not unique.

If this is for storage, like a file server and you do not have a sync process like rsync or DFS running to keep that replicated, you can look at corosync (http://corosync.github.io/corosync/)

Answer 2

If I have a processes running on s1 but then s1 fails somehow, how can I automatically run that process on one of the other 4 servers?

all major OS support clustering which does that - given that the process is installe on s2. And sometimes if the application is cluster aware. If that is not the case, virtualization can come to the rescue by starting the VM on another server.

Would that require shared storage across all these servers so the process can essentially just be started again on another machine?

That depends. If the process is locally installed on the other machines - as is normal for a cluster setup - then only if there are shared ressources it needs. Which may be the case for the majority of programs.

Example SQL Server, Exchange - will not need shared storage in some configurations because every cluster node keeps a copy of the data (because shared storage ALSO has issues - not so much on the hardware side which you can make redundant, but if a failing machine trashes your file system then ..... there is nothing to be shared ;)

If I have 5 IP addresses available for these 5 machines can I essentially assign those 5 just to the block of servers so each one can access each IP individually? So if a process runs and is accessed on xxx.xxx.xxx.1 but then s1 dies it swaps over and runs on s4, could s4 then have access to that IP address before so a user only sees a brief interruption in their service?

Any cluster system will do that. For windows, there is a cluster system integrated that allows you to assign an IP to a cluster program - which then moves when the program fails over.

SAdly you say nothing about oyur infrastructure. Anyhow, you got a generic answer - time to sit down and start reading documentation of whatever you have available on your operating system.