David Thompson:
|Randall,
|
|I've got all the patches applied, but now for the questions.
|
|1. When attaching to another server, must both servers have this
|codebase? Is it possible that comp1 has the new code, comp2 doesn't,
|but comp1 can still connect to comp2 via a new remote login (ssh)? If
|so, I can't get this to work on Linux. It dies around the select()
|function.
>> When attaching to another server, must both servers have this codebase? <<
No (at least not necessarily). I've been running a patched version here on
our SGIs to an unpatched (stock 4.1.0) DX on the IBM SP nodes.
The patches only affect how the invoking dxexec kicks off the remote
dxexec. However (caveat), I don't know all the internals of the execution
group code. If a slave dxexec can spawn other slave dxexecs, it would need
to be a patched one for the extra features (diff domain, user, env, etc.).
I haven't seen DX actually do this though; network data appears to be
routed through the master.
|2. Do you have time to complete this patch for the remote rsh macro
|code in DXChild and conn.c? It would be really nice for some people
|who wish to use ssh to connect to remote servers from a local UI and
|it would be nice if this worked consistently across the entire system.
Do you mean just consolidating the rsh references into one central spot in
the code? Or also making this setting user settable in a way besides the
environment variable?
I'd thought of standardizing the RSH #define to the one in
dx/include/dx/arch.h, but it's wrong and apparently not updated by
configure. I'm not an autoconf/autolocal expert or I'd have taken a look
at it.
Suggestion: What about just changing the arch.h #define RSH ref (for UNIX)
to:
#if defined(hp700) || defined(aviion)
# define RSH_CMD "/usr/bin/remsh"
#elif defined(sgi)
# define RSH_CMD "/usr/bsd/rsh"
#elif defined(sun4)
# define RSH_CMD "/usr/ucb/rsh"
#else
# define RSH_CMD "/usr/bin/rsh"
#endif
#define RSH ( getenv("DXRSH_ENV" ) ? getenv("DXRSH_ENV" ) : RSH_CMD )
Then we can just delete the RSH macro blocks in DXChild.c, conn.c, and
remote.c. That'd cover the local refs I believe, though I noticed there
are some explicit "rsh" refs in strings for remote commands. Possibly they
should also honor DXRSH_ENV, which (if set) should be set correctly in the
remote login profile (.cshrc/.profile/.bashrc).
|3. I would suggest using the environment variable DXRSH and
|DXRSH_NOENV since this is more similar to the existing environment
|variables.
That's fine with me. Feel free to update it.
Randy
--
Randall Hopper (mailto:[EMAIL PROTECTED])
Lockheed Martin Operation Support
EPA Scientific Visualization Center
US EPA MD/24 ERC-1A; RTP, NC 27711
