Matlab Parallel Computing - Firewall Configuration

Hi, I am not able to properly configure the firewall setting for the parallel computing cluster.

I have 2 servers. I installed mdce on both servers and added the mdce service to the firewall setting. The service runs fine on both servers.

However the mdce service is not reachable remotely, unless i turn off the firewall.

I get the following error when I run the function "nodestatus -remotehost server1"

java.lang.Exception: An error occurred while connecting to the mdce service on the host SERVER1.
        at com.mathworks.toolbox.distcomp.control.CommandSender.handleConnection
Exception(CommandSender.java:141)
        at com.mathworks.toolbox.distcomp.control.CommandSender.sendAndRunComman
d(CommandSender.java:59)
        at com.mathworks.toolbox.distcomp.control.CommandSender.sendAndRunComman
d(CommandSender.java:42)
        at com.mathworks.toolbox.distcomp.control.RunCommandSender.main(RunComma
ndSender.java:50)
Caused by: java.net.SocketTimeoutException: connect timed out
        at java.net.PlainSocketImpl.socketConnect(Native Method)
        at java.net.PlainSocketImpl.doConnect(Unknown Source)
        at java.net.PlainSocketImpl.connectToAddress(Unknown Source)
        at java.net.PlainSocketImpl.connect(Unknown Source)
        at java.net.SocksSocketImpl.connect(Unknown Source)
        at java.net.Socket.connect(Unknown Source)
        at com.mathworks.toolbox.distcomp.control.CommandSender.getSocketTryMany
Times(CommandSender.java:297)
        at com.mathworks.toolbox.distcomp.control.CommandSender.sendAndRunComman
d(CommandSender.java:54)
        ... 2 more
java.lang.Exception: An error occurred while connecting to the mdce service on the host server1.

The cause of this problem is:

============================================================================
   java.net.SocketTimeoutException: connect timed out
============================================================================

Script nodestatus unable to complete successfully - exiting

Any thoughts on how to fix this? I don't want to have the firewall turned off.

Answer 1

[Jason Ross]

Try out the steps in this solution, opening the ports as indicated. I've set it up several times on Windows and Linux and know it does work with a firewall.

http://www.mathworks.com/support/solutions/en/data/1-39LCCA/index.html?solution=1-39LCCA

Answer 2

[Jonny Coltraine]

Hi Jason,

Thanks for the answer, the problem above is resolved but now I am getting a new error:

ID # Start Time Stop Time Executing Node Subject Node Test Category Test Description Result Message 1 11-Sep-2012 20:24:07 11-Sep-2012 20:24:09 server1 server2 PORT_CONNECT PingServerSocketHost Test WARNING Host server2 does not respond to java.net.InetAddress.isReachable().

Both servers run window and contain the correct mapping of IP address and Hostname in the host file under windows/system32/drivers/etc/

Again, this issue does not appear when i turn off the firewalls.

thank in advance.

JC

[Jason Ross]

It sounds like you have a name resolution problem. There are ports on the firewall that are used for that, and it's likely cutting off that means of resolving the name when you turn it back on.

On Windows, try the following:

• In the System Properties dialog, you can set the computer name. Take a look at the "Full Computer Name" value on both the client and cluster nodes. Make sure they have what you expect them to have -- Windows can change this when you join a domain.
• In TCP/IP properties (you can get here off the properties for the network connection), in the Advanced ares, there is a "DNS" tab. Make sure that the DNS suffixes you need are appended.
• In the WINS tab, make sure the "Enable LMHOSTS lookup" is checked, and that the NetBIOS setting makes sense.
• Check that the ports are open on the firewall for name resolution to work. See this technet article for the port numbers:

http://technet.microsoft.com/en-us/library/cc738971(v=ws.10).aspx

[Jonny Coltraine]

HI Jason,

Still having same issue even thought the network seems to be configured well. Apart from the integrated matlab job manager, which other job scheduler would you recommend to use? I can see that matlab supports a few like Torque.

thanks JC

[Jason Ross]

MATLAB has direct integrations with Torque, PBS Pro, HPC Server 2008, LSF and SGE (through the generic interface). There's not one that we favor over the other, they can all run the various types of jobs, and the generic interface allows a few different filesystem layouts depending on how you do things.

I do know for a fact that the job manager runs fine with a firewall -- I've tested it with firewalls on for quite a few releases. Inevitably when I hit a problem it's because I've used the wrong ports from the client MATLAB to the cluster, as they do differ. But it does always work.

Answer 3

[Lee White]

I had also noticed Windows Firewall on Windows 7 64-bit was blocking my MDCS cluster from communicating. Specifically the admincenter listed my hosts as reachable but unavailable under status. I added the range of TCP ports 27350 to 29350 for incoming connections and the problem was solved.

Control Panel > Windows Firewall > Advanced Settings > incoming connection > new rule > allow TCP port range > Save

I would add some more documentation to the MDCS setup guides for other users will certainly experience this same problem. Furthermore, since the windows firewall configuration utility succeeded I was lead down the wrong path thinking it couldn't therefore be the firewall.

Matlab R2011a MDCS 5.1 Windows 7 64 bit across all computers.