Dear List,
I have a cluster of several PCs, each is running Fedora, either 14 or 15. I want to make this PCs grid enabled, so I installed globus ( http://www.globus.org/ ) and made a server with a global IP and with a subdomain. My main objective is to be able to run parallel programming (mpi), harnessing the idle CPUs of onlines nodes.
The stage at a moment, I have successfully set up simple-CA and the GSI of the server.
Globus commands works fine, for example:
++++ $ globus-hostname-lookup fedora.org Resolving address (fedora.org)... ------------------------------- Numeric address: 67.40.49.xxx Expected DN suffix: /CN=host/hoxx.trippy.org ++++
$ grid-proxy-init Your identity: /O=Grid/OU=GlobusTest/OU=simpleCA-localhost.localdomain/OU=localdomain/CN= Abu Attar Enter GRID pass phrase for this identity: Creating proxy ............................................ Done Your proxy is valid until: Wed Nov 2 05:01:04 2011
++++++
I tried but failed to copy a file from one node to a server.
$ globus-url-copy file:///home/username/Grid/Codes/cpi.c gsiftp://domainname/
The message is as follows,
error: globus_xio: Unable to connect to domainname:2811 globus_xio: System error in connect: No route to host globus_xio: A system call failed: No route to host
It seems something to do with the port 2811. Checking /etc/services of the server gave the port number and the protocol ++++ root$ less /etc/services | grep gsi
gsigatekeeper 2119/udp # GSIGATEKEEPER gsiftp 2811/tcp # GSI FTP gsiftp 2811/udp # GSI FTP gsidcap 22128/tcp # GSI dCache Access Protocol ++++
As I do not have any experience, I have no idea how to fix it. I mainly rely on internet searching, but still can not find the answer.
I expect from this list some suggestions or any comment. Thank you very much in advance.
with best, AA
On 11/01/2011 10:11 AM, Abu Attar Musharih wrote:
Dear List,
I have a cluster of several PCs, each is running Fedora, either 14 or 15. I want to make this PCs grid enabled, so I installed globus ( http://www.globus.org/ ) and made a server with a global IP and with a subdomain. My main objective is to be able to run parallel programming (mpi), harnessing the idle CPUs of onlines nodes.
Hi Abu Attar Musharih,
First of all, I don't think that a grid is the right instrument for running processes using a MPI model (this is more thread based, or, if using a cluster, based on a RPC model for example). Or look at *www.mosix.org*.
For cluster software, you could use Scientific linux or Rocks Clusters Linux. I made good experience in managing a cluster with Rocks Clusters Linux, based on CentOS (www.rocksclusters.org).
Additionally, in my opinion, the Fedora user list is not really a good contact for your grid question. I'm sure there is a appropriate Q&A page at globus.org (the messages listed below are not system specific enough) - it seems to be a middleware problem.
Kind regards
Joachim Backes
The stage at a moment, I have successfully set up simple-CA and the GSI of the server.
Globus commands works fine, for example:
++++ $ globus-hostname-lookup fedora.org Resolving address (fedora.org)...
Numeric address: 67.40.49.xxx Expected DN suffix: /CN=host/hoxx.trippy.org ++++
$ grid-proxy-init Your identity: /O=Grid/OU=GlobusTest/OU=simpleCA-localhost.localdomain/OU=localdomain/CN= Abu Attar Enter GRID pass phrase for this identity: Creating proxy ............................................ Done Your proxy is valid until: Wed Nov 2 05:01:04 2011
++++++
I tried but failed to copy a file from one node to a server.
$ globus-url-copy file:///home/username/Grid/Codes/cpi.c gsiftp://domainname/
The message is as follows,
error: globus_xio: Unable to connect to domainname:2811 globus_xio: System error in connect: No route to host globus_xio: A system call failed: No route to host
It seems something to do with the port 2811. Checking /etc/services of the server gave the port number and the protocol ++++ root$ less /etc/services | grep gsi
gsigatekeeper 2119/udp # GSIGATEKEEPER gsiftp 2811/tcp # GSI FTP gsiftp 2811/udp # GSI FTP gsidcap 22128/tcp # GSI dCache Access Protocol ++++
As I do not have any experience, I have no idea how to fix it. I mainly rely on internet searching, but still can not find the answer.
I expect from this list some suggestions or any comment. Thank you very much in advance.
with best, AA
On Tuesday 01 November 2011 09:11:01 Abu Attar Musharih wrote:
I have a cluster of several PCs, each is running Fedora, either 14 or 15. I want to make this PCs grid enabled, so I installed globus ( http://www.globus.org/ ) and made a server with a global IP and with a subdomain. My main objective is to be able to run parallel programming (mpi), harnessing the idle CPUs of onlines nodes.
Depending on what you actually want to do, globus may or may not be the right tool for the job. What exactly do you want to do, for what purpose, and what is your hardware setup?
I tried but failed to copy a file from one node to a server.
$ globus-url-copy file:///home/username/Grid/Codes/cpi.c gsiftp://domainname/
The message is as follows,
error: globus_xio: Unable to connect to domainname:2811 globus_xio: System error in connect: No route to host globus_xio: A system call failed: No route to host
It seems something to do with the port 2811.
This can be a number of things. However, in globus deployment (especially if you are doing it for the first time), the *number* *one* culprit is always the firewall. Globus uses *a* *lot* of custom ports, and it expects all of them to be available.
So, as a first step (and for testing purposes only!!!) take down the firewall on both machines (the client and the server), as
# service iptables stop
and retry the gsiftp. Be aware that shutting down the firewall is VERY DANGEOROUS and should be brought back up as soon as possible. However, when debugging Globus, it is the only reasonable way to check what is wrong.
If gsiftp is successful with firewall down, then do some research on what ports you need to open up and customize your firewall config. Expect to spend considerable time on that, since the config will be very, very custom. :-)
If it isn't the firewall, then there is something wrong with some part of your globus installation. It can be a number of things.
You are better off asking on one of the globus support mailing lists (I am not actively following any of those anymore, so cannot point you exactly). I doubt you will find many people with Globus experience on the Fedora list. ;-)
HTH, :-) Marko
On Tue, Nov 1, 2011 at 11:02 PM, Joachim Backes joachim.backes@rhrk.uni-kl.de wrote:
Hi Abu Attar Musharih,
First of all, I don't think that a grid is the right instrument for running processes using a MPI model (this is more thread based, or, if using a cluster, based on a RPC model for example). Or look at *www.mosix.org*. For cluster software, you could use Scientific linux or Rocks Clusters Linux. I made good experience in managing a cluster with Rocks Clusters Linux, based on CentOS (www.rocksclusters.org).
Thanks for the response. You are completely right, but for my main objective which is more on a proof-of-concept and learning the grid technology, perhaps this is the way to go.
Additionally, in my opinion, the Fedora user list is not really a good contact for your grid question. I'm sure there is a appropriate Q&A page at globus.org (the messages listed below are not system specific enough)
- it seems to be a middleware problem.
Seing that the globus toolkit is available from the fedora repo, I assume that there must be some members who can provide some explanation.
Best regards, Abu Attar
-
Abu Attar Musharih -
Joachim Backes -
Marko Vojinovic