[ofw] RE: WinOF 2.0 installation question

[Smith, Stan]

Hello,
  A couple questions and suggestions:

You do have a Mellanox ConnectX HCA in your system and not an older Mellanox InfiniHost HCA?

Download a new copy of ibclean.bat
>From a privileged administrator cmd window, run ibcleanup.bat

*** REBOOT ***

Once rebooted - What is the DeviceID for the Mellanox ConnectX HCA you have installed? Or do you know the model number?
run devmgmt.msc from a command window.
You should see under 'Other Devices' an InfiniBand Controller.
Select properties for the device.
Details Tab
Select 'Hardware IDs' properties
Email 1st 3 lines of Hardware IDs properties
cancel
file->exit

>From a privileged administrator cmd window, run ibscan.bat
Ibscan should show no lingering files and errs in not finding registry entries.
If this is the case, then

Install OpenFabrics WinOF 2.0.2  (WinOF_2-0-2_wlh_x64.msi) which you downloaded from http://www.openfabrics.org/downloads/WinOF/v2.0.2/Server_2008-Vista-HPC/Installers/

There is no reason I can think of as to why WinOF 2.0.2 should not install, except for lingering files and/or registry entries left over from previous installs.
The WinOF error your are seeing indicates the Mellanox ConnectX bus driver installed, although the installation of the InfiniBand HCA driver failed; not sure how this occurs as generally if one installs the other installs and both are required?

Try the ibcleanup.bat & reboot, 2.0.2 install and forward the HCA DeviceID before trying RC2.

thanks,

stan.






________________________________
From: Casano, Warenne [mailto:Warenne.Casano at kla-tencor.com]
Sent: Monday, July 20, 2009 11:27 AM
To: Smith, Stan
Cc: ofw at lists.openfabrics.org
Subject: RE: [ofw] RE: WinOF 2.0 installation question

Hi,

Thank you for your response.

I tried to reinstall WinOF again today.  Here is what I did:

(1) Manually removed "mlx4_* and ipoib*" from %windir%\system32\DriverStore\Temp.
(2) Reinstall WinOF 2.0
(3) Dialog prompting to install "Mellanox Technologies Ltd" device driver.  Select "install".
(4) Installation failed with the following error:

Mlx4_hca Install Err(2) cmd.exe /c cd /d C:\Program Files (x86)\WinOF\IBcore & devman.exe update mlx4_hca.inf PCI\VEN_15B3

Do you know what this error means?  I think I'll wait for WinOF 2.1 RC2 and try again.

Thanks
Warenne

________________________________
From: ofw-bounces at lists.openfabrics.org [mailto:ofw-bounces at lists.openfabrics.org] On Behalf Of Smith, Stan
Sent: Monday, July 20, 2009 10:30 AM
To: Casano, Warenne
Cc: ofw at lists.openfabrics.org
Subject: [ofw] RE: WinOF 2.0 installation question

Hello,
  Please see below.


________________________________
From: wwg-bounces at lists.openfabrics.org [mailto:wwg-bounces at lists.openfabrics.org] On Behalf Of Casano, Warenne
Sent: Monday, July 20, 2009 9:59 AM
To: wwg at lists.openfabrics.org
Subject: [Wwg] WinOF 2.0 installation question


Hi,

I am migrating an MPI application from Linux to Windows (Intel Xeon platform with
Windows 2008 server).  At first, I installed Mellanox WinOF 2.0 and tried to use it
with Intel MPI Library.  The Intel MPI library requires DAT runtime library which was
not part of the Mellanox WinOF 2.0 installation.  So, I uninstalled Mellanox WinOF 2.0.
I downloaded WinOF 2.0 and tried to install.  However, the installation failed.
>From the release notes, installation may fail if the certain files
are in DriverStore\Temp directory.  I could not removed those files even when I'm
logged in as "administrator".

I need to know how to remove those files from DriverStore\Temp directory.  I
tried using ibcleanup; but that did not remove the files in DriverStore\Temp.

<Stan> Latest version of ibcleanup.bat @  http://www.openfabrics.org/downloads/WinOF/etc/ know how to remove these files.
Are you sure the Mellanox WinOF 2.0 installation is removed? Check device manager (devmgr.msc)
 (You did reboot after Mellanox uninstall?)
Take a look at how ibcleanup.bat removes DriverStore folders if per-chance if does not work for you; perform task manually from privileged cmd window.

I am new in this area; so I would like to know how I could verify whether dapltest
works or not.  What I need is a sample output of when dapltest works after
WinOF is installed properly.

>From a head-node WinOF cmd window (Start->All Programs-->Windows OpenFabrics->Command Window
start a dapl2test server; type 'dt-svr<Enter>'
The dapl2test svr is listening on the IPv4 address you assigned to your IPoIB instance.
You will need that IPoIB IPv4 address in the next step; ipconfig will reveal if you're not sure.

>From another WinOF cmd window, type 'dt-cli IPoIB-IPv4-address conn' for a simple node local connection test; 'dt-cli' with no args gives help message.
Example: If 10.104.200 is assigned to your 1st IPoIB instance @ 'Local Area Connection 3' or in HPC speak 'Application Network'

    dt-svr<Enter>      # dapl2test server is listening on 10.10.4.200

    dt-cli 10.10.4.200 conn<Enter>
    dt-cli 10.10.4.200 transmet        # a more complicated test
    dt-cli 10.10.4.200 stop            # causes dap2test server to stop; otherwise it is persistent.

Now that you have proven Dapl2test is functional on the head-node, make sure the dapl2test server is still running on the head-node.
using clurun run dapltest clients from compute nodes; from the head-node command window type 'clusrun /nodes:n1,n2,n3 dt-cli 10.10.4.200 conn' for the 1st step.
Expand at your leisure.

Let me know how things go.

Stan.

PS: I you can wait until Tuesday 7/21 afternoon (PDT) WinOF 2.1 RC2 will be available - a much cleaner device install/uninstall! Strongly recommended.
PSS: wwg at lists.openfabrics.org<mailto:wwg at lists.openfabrics.org> is a small forum, I changed the reply to much larger audience of developers.

Thank you for your help.
Warenne
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ofw/attachments/20090720/f769fd75/attachment.html>