CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   CFX (http://www.cfd-online.com/Forums/cfx/)
-   -   About parallel processing in Linux (http://www.cfd-online.com/Forums/cfx/21501-about-parallel-processing-linux.html)

tuks August 5, 2005 04:33

About parallel processing in Linux
 
Hi friends,

I want to start the parallel computing on Linux,OS is Fedora core2, machine specifications are intel_p4.sse2_linux2.3.2, it is a 32bit processor.

In the documentation something is given about the "rsh unixhost echo working" command. Can you tell what does it mean by and how it is different than 'ssh'?

We have the parallel PVM computations license. We have installed the parallel processing license on the pc as server different than the server for CFX.

Can somebody tell what is the SETUP procedure for parallel processing to start?

While editing the 'hosts.ccl' (in cfxroot/cfx/config directory) file, do we need to add all the host computers which will be used for the parallel computing? I am sorry for asking lot's of questions but not much familier with Linux in detail.

Thanking you in advance.

Tuks

Ogbeni August 5, 2005 06:33

Re: About parallel processing in Linux
 
ssh is more secure than rsh, however it does not matter if you have your own local network,both will do the job. Only the master node need to have the names of the slave nodes in hosts.ccl, once you have setup the solver on the slave nodes, copy the host.ccl from each slave and put it in the one on the master node. rsh or ssh is funny in linux, I think you can not have a firewall for them to work, or you must list all the host id and ip addresses in the hosts.allow, the write up in the CFX manual is explicit. I wonder why you are using Fedora Core, which is explicitly not on the list for CFX5, I have a cluster of 4 Pentium 4 running CFX5.7 with Redhat 7.3, I have had no problems with it. I have another one on RH 9.0, and it took forever to get it to work. Good luck!

tuks August 5, 2005 06:56

Re: About parallel processing in Linux
 
Thanks Ogbeni, But i still have some doubts and some errors in solving the problem of starting the parallel run.

I think irrespective of Redhat/Fedora core, the procedure for setup for parallel running should be same.. And since i am new to it & do not know much details of Linux, i am facing some problem related to it.

I will get back to you with more information.

But still can you give me the procedure for setup, because i am not able to understand CFX manual for parallel running.

Thanks again for your quick reply.. hope for more

Tuks

Kalyan August 5, 2005 07:07

Re: About parallel processing in Linux
 
Hi U have to add all the machines in the Hosts.ccl file like one example below. HOST DEFINITION: 99ATXV5

Installation Root = V:\aeapps\CFX5\CFX-5.7.1\CFX-5.7.1

Host Architecture String = intel_p4.sse2_winnt5.0 END # HOST DEFINITION 99ATXV5 regards Kalyan

Ogbeni August 5, 2005 08:21

Re: About parallel processing in Linux
 
tuks, Lets try it this way: 1) Install CFX on all your nodes. 2) If you have not, assign names and IP address to all the nodes, eg the master nodes can be Master,IP=192.168.0.1, and the next 192.168.0.2 etc, (if you want to get on the internet you may need 2 ethernet cards, one for outside and one for local run) 3)If you will be using your own local network, you will need a switch, that is the traffic cop, you can get at the computer shop. 4)Connect your crossover or ethernet wires to the computers and then to the Switch 5) from any node, issue the command 'telnet name of computer or the ip address, if you get a response about login, that is good, if not you need to check your firewalls. 6) Now you can follow the instructions in the CFX manual. If you have done all this above, then be more specific with your questions. Good luck

Orca.Fight August 7, 2005 09:28

Why the rsh connecting refused?
 
Hi, all I use Fedora, and I turn off the Firewall and SELinux and reboot, but I still cannot rsh to localhost (even I cannot telnet localhost)? (note: I can ssh localhost) Can anybody help me ........T H A N K S !<font color=green> and more can anybody just give some example content of hosts.allow ,hosts.deny, hosts.equiv</font>


WT August 8, 2005 01:30

Re: Why the rsh connecting refused?
 
Hi,

I'm not familiar with Fedora, but the procedure is similarly tricky with SuSE 9.3 because in the default installation it also only supports ssh (due to security issues; btw, running in parallel is afaik also possible using ssh).

On SuSE, you have to install the rsh daemon on all of your machines which is not done by default. Then, you have to enable both the telnet and the rsh- daemon either by hand or by editing xinetd (or inetd, depending on which is installed). If these are running, telnetting and rshing should work for all machines, but probably with password. To remedy this, as pointed out above, the hosts.allow (or hosts.equiv) files have to be put onto all machines. Now logging in without password should work and you're set to start.

Hth

-Thomas

Tuks August 8, 2005 02:24

Re: About parallel processing in Linux
 
Hi Ogbeni,

Thanks for your advice but still i am not able to resolve the problem.. I had added the hosts " CFD-3 and CFD-2" into the 'hosts.ccl' file and the usrename on all the computers is same "usr" and CFX is installed in same directory on each computer.. still i am getting following ERROR.

"Warning! Host name lookup failed for host CFD-3

Warning! Host name lookup failed for host CFD-2

An error has occurred in cfx5solve: Remote connection to cfd3 (CFD-3) could not be started, or exited with return code 255.

It gave the following output:

ssh: CFD-3: Name or service not known

Check that you have typed the hostname correctly, that you have an account "user" on the specified host with permission to rsh from this host."

It seems to be simple problem, related to the hosts recognition by the master computer.. but i checked the hostnames they are same what i mentioned.

What could be the problem? Can you please help me.. I have tried to follow all the steps given in the CFX_MANUAL but couldn't help.. May be because i am not much familier about LINUX..

Again Thanks in advance..

Tuks

Orca.Fight August 8, 2005 02:43

Re: About parallel processing in Linux
 
Thank to WT! I'll try it.


To Tuks:
You need to edit /etc/hosts and it looks like:
192.3.4.5 CFX-3
192.4.5.6 CFX-2

Ogbeni August 8, 2005 06:35

Re: About parallel processing in Linux
 
Hello Orca is right, please also check that you have the same user name on all your nodes. For instance if your user name on master node is CFD, and CFX is installed in /home/CFD/CFX-5.7, then all your nodes must have CFX set up in the same way, in the same directory name. Also do not set it up as root.

tuks August 8, 2005 08:22

Re: About parallel processing in Linux
 
Hi all, Thank you for your kind advice. I have been able to find out the problem and were able to sort it. Actually problem was with the recognition of host machines. I donot have the local nodes but have different machines in network, for which i need to edit ssh 'authentication_keys' and permit for SSH without asking for passward everytime when ever you ascess/exchange the data between the machines.. which can be done by GENERATING THE KEY PAIRS.

This problem was sort by using the "ssh-keygen -t dsa" command which creates the 'id_dsa.pub' file and which must be copied to the other machine '~/.ssh/authorized_keys2'

The procedure is given online

"http://www.redhat.com/docs/manuals/linux/RHL-7.2-Manual/custom-guide/openssh-clients.html"

Thank you again for your kind help. I hope to have very good inteaction in future as well.

With regards tuks


All times are GMT -4. The time now is 13:07.