LAE

Note: This was originally posted by an inactive account. Content was preserved by moving under an admin account.

Originally posted by: stevem

In the Graph specific configuration files how would I set the server farm up for more threads.

In 4.6 we have;

export p_ls_brain_controller_farm='("server.company.com:$ {ls.brain.server.port}:16")'

How would I set it up when we will have 2 servers and we want to share the load. i.e. run the threads evenly between them.

Note: This was originally posted by an inactive account. Content was preserved by moving under an admin account.

Originally posted by: stonysmith

It's not really possible to run them "evenly", but to split across multiple servers you can do this:

export p_ls_brain_controller_farm='("server1:$ {ls.brain.server.port}:4","server2:$ {ls.brain.server.port}:4","server1:$ {ls.brain.server.port}:4","server2:$ {ls.brain.server.port}:4")'

You could obviously do it this way:
export p_ls_brain_controller_farm='("server1:$ {ls.brain.server.port}:1","server2:$ {ls.brain.server.port}:1","server1:$ {ls.brain.server.port}:1","server2:$ {ls.brain.server.port}:1"..............)'
but that is a real mess to maintain.

It is important to note that the servers MUST be able to read each other's temp files. You will have to work a bit to get the temp file folder names setup properly, and it is fairly common that you may want the standard file access to be "SFTP" instead of "FILE"... it all depends on your disk mounts.

Note: I myself would use this shortcut:
export s1="server1:port1:1"
export s2="server2:port2:1"
export p_ls_brain_controller_farm='("${s1}","${s2}","${s1 }","${s2}",..............)'


=======================
A bit of explanation on load balancing. Let's say that you have four nodes in a sequence and that nodes A and C run very quickly, but nodes B and D are very long running. In the configuration above, nodes A and C will run on server 1, nodes B and D will run on server2.
In that configuration, the two long running nodes will both run on server2, leaving server1 basically idle. There is some work with thread pooling that you can use to affect this (read the documentation) but achieving "true" load balancing is going to be tricky.