On Wed, Feb 16, 2011 at 3:55 PM, Jeffrey Hutzelman <[email protected]> wrote:
> --On Wednesday, February 16, 2011 09:14:55 AM -0600 Andrew Deason
> <[email protected]> wrote:
>
>> On Tue, 15 Feb 2011 22:59:23 -0500
>> Jeffrey Hutzelman <[email protected]> wrote:
>>
>>> On SPARC, there is virtually no advantage to building 64-bit versions
>>> of most binaries,
>>
>> Virtual memory space?
>
> _Most_ binaries don't need more than 2GB of virtual address space.
> The fileserver would likely benefit, on machines with enough RAM to make
> this worthwhile.  Most of our programs certainly would not.
>

Hi Jeff,

The new DAFS threaded salvager additionally falls into this category
(as well as the standalone salvager when large volumes are involved,
thanks to the--potentially enormous--vnode essence vectors).
Furthermore, fssync-debug and salvsync-debug need to be 64-bit
binaries as well (thanks to us directly marshalling pointers and other
variable-width types, which is, admittedly, a design flaw for which
I'll take responsibility).  To simplify the build system (and avoid a
few fssync protocol changes), this also means volserver should be a
64-bit binary...

> I'd suggest building a 64-bit fileserver and 64-bit versions of most of the
> libraries we export (in addition to the 32-bit versions, of course).  But
> that's about it.
>

I concur this would be ideal.  However, then we get into the argument
about whether our build system has any business twiddling CFLAGS.  I
suppose we could have a special configure switch to do this special
behavior for people who want an optimal build, but that is likely to
be quite contentious...

> Of course, that needs to be selectable, but I think it's a fine default.
> There are very few SPARC Solaris machines that are likely to run current
> OpenAFS but are not 64-bit capable.  I'm not sure current Solaris still even
> supports anything that old.

32-bit (sparc) kernel support was dropped during development of
Solaris 10.  For how much longer do we want to support 8 and 9?

-Tom
_______________________________________________
OpenAFS-devel mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-devel

Reply via email to