So, I''ve been charged with testing 1.6 (well, 1.5.97) with the Cray XT3, but I seem to be running into a problem. Utilizing lustre 1.4.7 based patches, I can successfully build and run my own kernel outside of the official Cray build process (if this seems like a trivial sentence, you don''t have the pleasure of using an XT3). The problem is when I go back and utilize the lustre 1.5.97 patch set and build a kernel. It boots just fine, but the ethernet devices (e1000 based) do not behave (for instance, I can ping the node just fine, and it can mount NFS mounts just fine, but if I attempt to access those NFS mounts the process will hang indefinitely). So, with this, I''m about to start the journey of looking through the 1.5.97 patches to see what might be giving me grief, but I was hoping that someone smarter than I might have some insight on some things to try. Thanks... -- Makia Minich <minich@ornl.gov> National Center for Computation Science Oak Ridge National Laboratory --*-- Imagine no possessions I wonder if you can - John Lennon
There should actually not be any difference between the patches between a late 1.4.X and 1.6 -- we''re trying to keep the kernel patches in sync. So it''s likely that you can just use a 1.4.x-patched kernel you already have working, and build 1.6 against it. Makia Minich wrote:> So, I''ve been charged with testing 1.6 (well, 1.5.97) with the Cray XT3, but I > seem to be running into a problem. Utilizing lustre 1.4.7 based patches, I > can successfully build and run my own kernel outside of the official Cray > build process (if this seems like a trivial sentence, you don''t have the > pleasure of using an XT3). The problem is when I go back and utilize the > lustre 1.5.97 patch set and build a kernel. It boots just fine, but the > ethernet devices (e1000 based) do not behave (for instance, I can ping the > node just fine, and it can mount NFS mounts just fine, but if I attempt to > access those NFS mounts the process will hang indefinitely). > > So, with this, I''m about to start the journey of looking through the 1.5.97 > patches to see what might be giving me grief, but I was hoping that someone > smarter than I might have some insight on some things to try. > > Thanks... >
Can you be more specific on what is considered a "late 1.4.X" version? On Wednesday 25 April 2007 5:25:38 pm Nathaniel Rutman wrote:> There should actually not be any difference between the patches between > a late 1.4.X and 1.6 -- we''re trying to keep the kernel patches in > sync. So it''s likely that you can just use a 1.4.x-patched kernel you > already have working, and build 1.6 against it. > > Makia Minich wrote: > > So, I''ve been charged with testing 1.6 (well, 1.5.97) with the Cray XT3, > > but I seem to be running into a problem. Utilizing lustre 1.4.7 based > > patches, I can successfully build and run my own kernel outside of the > > official Cray build process (if this seems like a trivial sentence, you > > don''t have the pleasure of using an XT3). The problem is when I go back > > and utilize the lustre 1.5.97 patch set and build a kernel. It boots > > just fine, but the ethernet devices (e1000 based) do not behave (for > > instance, I can ping the node just fine, and it can mount NFS mounts just > > fine, but if I attempt to access those NFS mounts the process will hang > > indefinitely). > > > > So, with this, I''m about to start the journey of looking through the > > 1.5.97 patches to see what might be giving me grief, but I was hoping > > that someone smarter than I might have some insight on some things to > > try. > > > > Thanks...-- Makia Minich <minich@ornl.gov> National Center for Computation Science Oak Ridge National Laboratory Phone: 865.574.7460 --*-- Imagine no possessions I wonder if you can - John Lennon
Makia Minich wrote:> Can you be more specific on what is considered a "late 1.4.X" version? >A version that came out around the same time the beta you''re using came out. In practice, much of the time our kernel patches and the Lustre versions that work with them are quite forgiving. You might just try building 1.5.97 against your 1.4.7-patched kernel -- the Lustre configure script may very well do the right thing.> On Wednesday 25 April 2007 5:25:38 pm Nathaniel Rutman wrote: > >> There should actually not be any difference between the patches between >> a late 1.4.X and 1.6 -- we''re trying to keep the kernel patches in >> sync. So it''s likely that you can just use a 1.4.x-patched kernel you >> already have working, and build 1.6 against it. >> >> Makia Minich wrote: >> >>> So, I''ve been charged with testing 1.6 (well, 1.5.97) with the Cray XT3, >>> but I seem to be running into a problem. Utilizing lustre 1.4.7 based >>> patches, I can successfully build and run my own kernel outside of the >>> official Cray build process (if this seems like a trivial sentence, you >>> don''t have the pleasure of using an XT3). The problem is when I go back >>> and utilize the lustre 1.5.97 patch set and build a kernel. It boots >>> just fine, but the ethernet devices (e1000 based) do not behave (for >>> instance, I can ping the node just fine, and it can mount NFS mounts just >>> fine, but if I attempt to access those NFS mounts the process will hang >>> indefinitely). >>> >>> So, with this, I''m about to start the journey of looking through the >>> 1.5.97 patches to see what might be giving me grief, but I was hoping >>> that someone smarter than I might have some insight on some things to >>> try. >>> >>> Thanks... >>> > >