General

• On all platforms, you should set things up such that the xrootd and olbd daemons are restarted automatically should they crash for some reason. While the servers are fairly stable, there is always the possibility of some rare crash may happen. Due to the fault-tolerance, the clients themselves will not crash and can continue as long as the servers are restarted. Typically this can be done with xinetd or with a simple cron job that checks every 10 minutes (10 minutes is within the time that the clients will loop, waiting for the server to return).

Linux/x86

• The hard limit for number of open file descriptors is typically set to 1024 in linux distributions. If you expect to have many clients connecting to your xrootd/olbd system, you should increase this on both the data servers and the redirector (it is more important for the redirector). A common value to which people change this is 16384.

Solaris

• Due to a feature of how the virtual memory is managed in Solaris you should allocate more swap space then you might normally for a server: for a typical server with 2GB real memory we recommend allocating 16GB of swap space. (This large swap isn't necessary on Linux, for example, as it manages the virtual memory more economically.)