OmniRPC User's Guide Version 1.2: http://www.omni.hpcc.jp/omnirpc/ <omnirpc@omni.hpcc.jp> | ||
|---|---|---|
| Prev | ||
A: In the OmniRPC system, after the agent's invocation, the program on the remote host requests access to the client host. Therefore, it is necessary for the remote host to know the hostname which can be accessed to the client host By default, the OmniRPC system uses the hostname through hostname commands. But, in some settings, there may exits a hostname which cannot be accessed from an outside network exists. You should set FQDN(Full Qualified Domain Name) by the hostname command or environment variable OMRPC_HOSTNAME as FDQN.
(csh or tcsh)
% setenv OMRP_HOSTNAME FQDN
(bash)
$ export OMRP_HOSTNAME FQDN
There is the same problem in the Globus Toolkit environment. In this case, set the environmental variable GLOBUS_HOSTNAME to FQDN. for more details, please see Globus information.
A: Usually cluster nodes accept RSH, So, the agent can invoke the remote executable program on the cluster nodes. But in some situations, the cluster nodes restrict the use of RSH and support SSH. Therefore, you can use SSH to invoke remote executable programs. You should write explicitly to use SSH. You can change the cluster nodes file (which is introduced in Use of a built-in round-robiin scheduler) as below.
hpc1 ssh
hpc2 ssh
hpc3 ssh
If ssh is omitted in the above description, rsh is used.
Q: I use SSH both for agent invocation and for worker invocation when I want to use clusters. But An error has occured when invocation of workers.
A: It seems that authentications between OmniRPC agent and worker program are failed. Because OmniRPC agent cannot use the ssh-agent's pass-phrase in which client program runs. As result, when OmniRPC agent uses ssh to invoke worker program in cluster nodes, authentication between agent and worker program is failed. Easy way to solve this issue is to add ssh option in ".ssh/config" as follows.
ForwardAgent yes
Q: After I lunched the client program, client program exits with like bellow message.
OMRPC_FATAL(localhost:./[programname]): omrpc_io_accept: time out
A: Authentication phase sometimes takes more than 15 seconds when client program invokes OmniRPC agent in remote nodes. OmniRPC's default timeout is 15 seconds. If you want increase this number, please set the TimeOut element in hostfile like bellow example.
<?xml version="1.0" ?> <OmniRpcConfig> <Host name="jones.is.tsukuba.ac.jp"> <TimeOut second="20"> </OmniRpcConfig>