I'm unable to start a remote kernel, and have not found a solution to my problem the archives. I'm running Mathematica 7.0 on Windows Vista, trying to start a remote kernel on a Linux server also running 7.0.
I'm relying on WolframSSH.jar to make the connection, so I haven't fiddled with the advanced settings under kernel configuration options. (Well, I've tried without success to use putty to make the connection). When I evaluate the first command in a notebook using the remote kernel, I get the authentication key question and then enter a password, and then nothing happens. The notebook claims to be running but never evaluates, there's no error message, and Mathematica locks up.
I can see that a MathKernel does start on the remote machine, so the login at least is successful. And force closing Mathematica locally kills the remote kernel.
The only clue I can discover is that Mathematica wants to put the RSA key in the wrong place. My username on both machines is "carroll" (which I can confirm Mathematica knows with $UserName), but it creates the known_hosts file in "C:\Users\Ian\.ssh\" which wouldn't otherwise exist. I also don't see any key created on the remote machine. Could this cause the problem? What else should I try doing?
Thanks -Ian (* apologies if this is a double post, the first vanished *)
I have followed the very simple instructions for setting up a remote kernel in Mathematica 7.0 on Windows Vista, but my connection is not working. I believe the source of the problem is the location of the ssh keys. After telling a notebook to start the remote kernel, I get to a dialog which says:
"C:\Users\Ian\.ssh\known hosts does not exist Are you sure you want to create it?"
My username is "carroll" not "Ian", and Mathematica knows this:
In = $UserName Out = carroll
"carroll" is also my username on the remote machine. How do I force Mathematica to create and look for the known_hosts file in "C:\Users \carroll\.ssh\"?
And yes, even if I let the known_hosts file get created in Users\Ian \.ssh, the connection still does not work.
Thanks to help from bj...@mit, but I'm still not there. I'm now getting a msg "The kernel cruncher failed to connect to the front end. (Error = MLECONNECT). You should try running the kernel connection outside the front end." I tried, but what sort of behavior is good or bad?
My configuration on a Windows Vista machine running Mathematica 7.0 is
Arguments to MLOpen: -LinkMode Listen -LinkProtocol TCPIP -LinkOptions MLDontInteract
When I run exactly this from cmd line in Vista, I get
> bash: linkname: command not found > bash: ipaddress: command not found > LinkOpen::linke: --Message text not found -- (MathLink linkname missing or malformed.)
When from the cmd line in Vista I run plink -i keyfile carroll@cruncher -x math -mathlink -linkmode Connect - linkprotocol TCPIP -linkname 5...@local.machine
there's no message, and I have to ctrl-C to get back to the cmd prompt. a MathKernel does start on cruncher. should something else happen?
thanks for any help! -Ian
On Nov 5, 1:24 am, Ian <carroll....@gmail.com> wrote:
> I have followed the very simple instructions for setting up a remote > kernel in Mathematica 7.0 on Windows Vista, but my connection is not > working. I believe the source of the problem is the location of the > ssh keys. After telling a notebook to start the remote kernel, I get > to a dialog which says:
> "C:\Users\Ian\.ssh\known hosts does not exist Are you sure you want to > create it?"
> My username is "carroll" not "Ian", and Mathematica knows this:
> In = $UserName > Out = carroll
> "carroll" is also my username on the remote machine. How do I force > Mathematica to create and look for the known_hosts file in "C:\Users > \carroll\.ssh\"?
> And yes, even if I let the known_hosts file get created in Users\Ian > \.ssh, the connection still does not work.
I've never used the WolframSSH.jar utility, but I have used plink to do this. Instructions are below (it sounds like you've done some of these steps). -Zach
1. Enable sshd on the slave. At a terminal, type # sudo apt-get install openssh-server openssh-client
2. Open Mathematica and go to Evaluation>Parallel Kernel Configuration. Click on the Remote Kernels tab. Click "Add Host."
3. Type the hostname or IP address of the slave (both will work) in the Hostname box.
4. Select "Use custom launch command" and change it to this: C:\Progra~2\plink.exe usern...@192.168.0.1 -pw 123456 -x "math -mathlink -linkmode Connect `4` -linkname `2` -subkernel -noinit >& /dev/null &"
where C:\Progra~2\plink.exe is the location of plink.exe (it might be Progra~1 if you're not on a 64-bit machine), username is your username on the slave, 192.168.0.1 is the hostname or IP address of the slave and 123456 is the password for your login on the slave.
5. Restart Mathematica. (This is not really necessary, you can just evaluate LaunchKernels[] but sometimes Mathematica bawks at your already-open kernels.) All of your kernels should launch. Test it with e.g. ParallelEvaluate[2^2048;//Timing] to see how your kernel speeds roughly compare.
If you get an error about your remote kernel(s) not launching, try running the command from step 4 from a command shell. (Open cmd.exe and type C:\Progra~2\plink.exe usern...@192.168.0.1 -pw 123456 "math" with the replacements described above. You should get In[1]:= in a shell.) If that doens't work, either PuTTY or your slave isn't configured properly.
> I'm unable to start a remote kernel, and have not found a solution to > my problem the archives. I'm running Mathematica 7.0 on Windows Vista, > trying to start a remote kernel on a Linux server also running 7.0.
> I'm relying on WolframSSH.jar to make the connection, so I haven't > fiddled with the advanced settings under kernel configuration > options. (Well, I've tried without success to use putty to make the > connection). When I evaluate the first command in a notebook using the > remote kernel, I get the authentication key question and then enter a > password, and then nothing happens. The notebook claims to be running > but never evaluates, there's no error message, and Mathematica locks > up.
> I can see that a MathKernel does start on the remote machine, so the > login at least is successful. And force closing Mathematica locally > kills the remote kernel.
> The only clue I can discover is that Mathematica wants to put the RSA > key in the wrong place. My username on both machines is > "carroll" (which I can confirm Mathematica knows with $UserName), but > it creates the known_hosts file in "C:\Users\Ian\.ssh\" which wouldn't > otherwise exist. I also don't see any key created on the remote > machine. Could this cause the problem? What else should I try doing?
> Thanks > -Ian > (* apologies if this is a double post, the first vanished *)
You say that the MathKernel is starting on the remote machine, so the ssh machinery worked. Don't worry about where known_hosts got created...there may be a minor issue here, but it's not causing your principal problem. And neither WolframSSH nor ssh itself will create a key on another machine. There are, of course, options for you to do this manually and for these ssh implementations to use that key...but they don't work by creating anything on the remote system. There seems to be a little bit of confusion in general about ssh here, but as I said, let's set all of that aside. You have confirmed (and I'll trust you did this via reliable means) that MathKernel was launched, and that means the ssh hurdle was cleared.
So, what are the other possible problems? These are the ones I can think of...
* The command-line might not identify the machine running the FE correctly. If you can get the command-line which was run on the remote machine, check to see how it identified your machine, and then ping that location. Did it work?
* You might have multiple network adapters on the FE machine, and the kernel might not be connecting to the same adapter that the FE is listening on.
* There might be an intervening firewall blocking traffic on arbitrary ports.
The first two can be fixed by fiddling with the command-line which is sent via ssh. The third, I'm afraid, is a real problem, for which there is no good solution.
Also, if you talk to Wolfram Tech Support, they can walk you through a process of manually launching an FE on one side, a kernel on the other, and hooking them up step by step in a way that will hopefully produce better feedback to what the real problem is (I could do this, too, but it'd about triple the length of this email, and I'm really very, very busy right now).
Sincerely,
John Fultz jfu...@wolfram.com User Interface Group Wolfram Research, Inc.
On Thu, 5 Nov 2009 03:52:05 -0500 (EST), Ian wrote: > I'm unable to start a remote kernel, and have not found a solution to > my problem the archives. I'm running Mathematica 7.0 on Windows Vista, > trying to start a remote kernel on a Linux server also running 7.0.
> I'm relying on WolframSSH.jar to make the connection, so I haven't > fiddled with the advanced settings under kernel configuration > options. (Well, I've tried without success to use putty to make the > connection). When I evaluate the first command in a notebook using the > remote kernel, I get the authentication key question and then enter a > password, and then nothing happens. The notebook claims to be running > but never evaluates, there's no error message, and Mathematica locks > up.
> I can see that a MathKernel does start on the remote machine, so the > login at least is successful. And force closing Mathematica locally > kills the remote kernel.
> The only clue I can discover is that Mathematica wants to put the RSA > key in the wrong place. My username on both machines is > "carroll" (which I can confirm Mathematica knows with $UserName), but > it creates the known_hosts file in "C:\Users\Ian\.ssh\" which wouldn't > otherwise exist. I also don't see any key created on the remote > machine. Could this cause the problem? What else should I try doing?
> Thanks > -Ian > (* apologies if this is a double post, the first vanished *)
It sounds like you're doing more than is necessary, but I might be misunderstanding your e-mail. My replies are interspersed below.
On 11/6/2009 5:15 AM, Ian wrote:
> Thanks to help from bj...@mit, but I'm still not there. I'm now > getting a msg "The kernel cruncher failed to connect to the front end. > (Error = MLECONNECT). You should try running the kernel connection > outside the front end." I tried, but what sort of behavior is good or > bad?
> My configuration on a Windows Vista machine running Mathematica 7.0 is
Try using the *exact* ML string that I sent in my instructions. The linkmode needs to be connect; you need to specify a host name, ML needs to interact, etc. This looks more like a slave string, but I don't know where you got it.
> When I run exactly this from cmd line in Vista, I get
>> bash: linkname: command not found >> bash: ipaddress: command not found
You shouldn't be running anything from the Windows shell except for diagnostics. Mmeca launches the remote kernel on its own. The syntax you wrote is specific to Mmeca as well (the `linkname` part, at least)
>> LinkOpen::linke: --Message text not found -- (MathLink linkname missing or malformed.)
> When from the cmd line in Vista I run > plink -i keyfile carroll@cruncher -x math -mathlink -linkmode Connect - > linkprotocol TCPIP -linkname 5...@local.machine
> there's no message, and I have to ctrl-C to get back to the cmd > prompt. a MathKernel does start on cruncher. should something else > happen?
If your connection is successful from a command line (e.g. if you "ssh yourserver math" you should get a Mathematica command line interface ">" and be able to e.g. type 1+1 and get 2).
> thanks for any help! > -Ian
Try following my instructions exactly; I think that will clear up your problem. You don't need to do anything outside of Mmeca to use the remote kernel after you set it up. Cheers, Zach
> On Nov 5, 1:24 am, Ian<carroll....@gmail.com> wrote:
>> I have followed the very simple instructions for setting up a remote >> kernel in Mathematica 7.0 on Windows Vista, but my connection is not >> working. I believe the source of the problem is the location of the >> ssh keys. After telling a notebook to start the remote kernel, I get >> to a dialog which says:
>> "C:\Users\Ian\.ssh\known hosts does not exist Are you sure you want to >> create it?"
>> My username is "carroll" not "Ian", and Mathematica knows this:
>> In = $UserName >> Out = carroll
>> "carroll" is also my username on the remote machine. How do I force >> Mathematica to create and look for the known_hosts file in "C:\Users >> \carroll\.ssh\"?
>> And yes, even if I let the known_hosts file get created in Users\Ian >> \.ssh, the connection still does not work.
We use a lot of remote kernels at work, and from home to work. The configurations we have are Mac to Mac, Mac to Linux, Mac to Windows and Windows to Windows. Had trouble with them all until we found this:
> I'm unable to start a remote kernel, and have not found a solution to > my problem the archives. I'm running Mathematica 7.0 on Windows Vista, > trying to start a remote kernel on a Linux server also running 7.0.
> I'm relying on WolframSSH.jar to make the connection, so I haven't > fiddled with the advanced settings under kernel configuration > options. (Well, I've tried without success to use putty to make the > connection). When I evaluate the first command in a notebook using the > remote kernel, I get the authentication key question and then enter a > password, and then nothing happens. The notebook claims to be running > but never evaluates, there's no error message, and Mathematica locks > up.
> I can see that a MathKernel does start on the remote machine, so the > login at least is successful. And force closing Mathematica locally > kills the remote kernel.
> The only clue I can discover is that Mathematica wants to put the RSA > key in the wrong place. My username on both machines is > "carroll" (which I can confirm Mathematica knows with $UserName), but > it creates the known_hosts file in "C:\Users\Ian\.ssh\" which wouldn't > otherwise exist. I also don't see any key created on the remote > machine. Could this cause the problem? What else should I try doing?
> Thanks > -Ian > (* apologies if this is a double post, the first vanished *)
|
