Connecting the server

[peterchu]

I have four unix-type servers in the network , they all have different application and in the same subnet and have static IP address eg. the IP address is 192.168.0.1 , 192.168.0.2 , 192.168.0.3 , 192.168.0.4 . Now , if I want to make all these servers can have fail-over function to make sure to can connect to the network , is it possible ? for example , a user want to connect 192.168.0.1 , if this server is corrupted suddenly , then the user will automatically login another server 192.168.0.2 , if this server is corrupted also , then switch to another server , of course they have the same password in all these servers, is this possible ? how to make it ? thx in advance.

[Jaime Bolanos Rojas.]

Peterchu,

That will depend on the applications that you are running, if it's DNS, you usually configured a master server and slave server, same for nis and nfs, you tell the client that if one server is not available, to connect to the other server.

Clusters they kind of work the same way.
Since the 4 boxes are running different unix type servers, then it might be a little bit different, please specify applications that you are running in those servers that are so important,

Regards,

Jaime.

[peterchu]

thx Jaime Bolanos Rojas ,

For the DNS method , can advise what setting I need to make in the unix box ? what file I need to update ? thx in advance.

[A. Clay Stephenson]

You've already indicated that these 4 servers are running different applications so it makes no sense, for example, if Oracle (non-RAC) is running on Server A to expect Server B to be able to run the same instance of Oracle. What you are describing is best handled by MC/Service Guard. Under MC/SG, the users don't connect to ServerA or ServerB but instead connect to "packages" which have assigned IP addresses that move to whatever host is actually running the package. Do a little searching under docs.hp.com for Service Guard.

Password sharing is done via NIS, NIS+, and/or LDAP.

[nash11]

hi A. Clay Stephenson ,

whether Service Guard is a package or not ? if use DNS method , is it more simple ? thx.

[A. Clay Stephenson]

DNS really has nothing to do with this. The problem is that 4 different applications are running on four different servers? What can DNS do in that case? Now, if all 4 applications were running on all 4 hosts AND somehow the data were kept in sync between all 4 hosts then DNS could be used -- the trick would be that you would would make DNS entries for something like "mickey.mouse.com" that would round-robin among all 4 servers and you would have daemons which would detect when server A stop responding and would make a dynamic DNS update to remove server A from the round-robin. Confused yet? I actually do something like this for multiple application servers that connect to a common database under MC/SG.

In any evnt, MC/SG is the correct approach for this problem but it is neither simple to properly setup or cheap.

[nash11]

thx A. Clay Stephenson ,

does MC/SG need to pay $ to purchase it ? is there any free solution for it ? does windows DNS can fit my requirement ? thx

[A. Clay Stephenson]

MC/SG is not free. Since you haven't described YOUR problem I have no idea if DNS (Windows or otherwise) is the answer.

[Steven E. Protter]

Shalom,

DNS has built in, free high availability functionality. You set up a master slave system and you are providing DNS services regardless of whether the master is running or not.

SG runs about $1000 or so per CPU used.

The model for its high availability is:

1) Data for the application must be available on shared storage for a multi node setup(it is possible but usess to set it up single node).
2) The application binaries can be on shared storage but realistically should be duplicated on two nodes.
3) A configuration is generated by you to manage high availability functions, which node is running, what happens if a node fails, avoiding two nodes trying to be primary and such.
4) Users connection is generally to a floating or shared ip address that moves from node to node depending on whats active.

Realistically, you should identify critical sevices and decide which ones you need higher availability on.

Sendmail combined with proper DNS configuration also provides high availability because it checks the MX mail exchanger records when trying to deliver mail. If the top record points to an unavailable or too busy server, it moves on to the next server.

SEP

[nash11]

thx Steven E. Protter ,

that is want I want , but except use the SG , how can I do that ? thx