Fw: To amd64 or not to amd64?

[Gabriele Cecchetti]

I add myself to this thread for a simple question.
We use amd64 even on Conroe 2 when we use our PC as Internet servers.
Maybe is wrong, but this sounds as 'a little harder' to attack (maybe 
some known attacks works on i386 code but not on amd64 code).
The main drawback I encounter (or I think so), it is with some contrib 
software that does not work as we wish, or is not available at all.
For example we would like to use CODA, but the client side does not work 
(venus panics as documented by a PR), we  would like to use OpenAFS, but
we did not find any working distribution (while is available for i386), 
ggate right now does not work for another documented problem.
I know that i386 platform has a lot of open PRs, but the feeling is that
with amd64, if you play hard... the things will get harder! :-)

_IMHO_

Gabriele

Mike Meyer ha scritto:
> In <200610111149.k9BBnaxm001133 at peedub.jennejohn.org>, Gary Jennejohn <garyj at jennejohn.org> typed:
>> "alecn2002" writes:
>>> The question is: will I have any benefits if I'll move to amd64 system, or it
>>> 's safer and better to stay with i386 arch?
>> [snip]
>>> Problem-free operation and stability have precedense for me over system speed
>> The only real advantage of 64-bit over 32-bit is that you can put more than
>> 4 GB of RAM into the box.
> 
> Personally, I think the 30% performance improvment I see using amd64
> is a real benefit. The amd64 architecture has twice as many registers
> as i386, and they're twice as wide. If you can run code built for
> amd64, you win. If you're running mostly i386 apps (as with certain
> proprietary systems), you don't.
> 
> It took a little bit of doing, but I got everything I used on my
> desktop on i386 working on amd64. The biggest headach was openoffice,
> but that was because one of the repositories had a broken tarball;
> once I got the right tarball, it built just fine. Two ports required
> tweaking to get the code to build on amd64. I wound up installing
> another port from the package. The other 400+ ports I use all worked
> with no problems. That 1% problem rate sounds about right for i386
> ports as well.
> 
> 32 bit emulation works, and as others have said, is how you run Linux
> binaries. The linuxolator works fine. In fact, that may be the best
> way to use 32 bit software. Whlle you can run 32 bit freebsd binaries,
> the ports system doesn't work for them, because there's no way to
> specify that you need the 32-bit version of a library, not the 64-bit
> version (if there is, please point me at the docs), so you don't get
> the library dependencies installed properly.
> 
> The other gotcha is that, while gcc knows how to build 32 bit
> binaries, the system compiler isn't configured to do the right thing
> by default. It's probably possible to make this work right, or maybe a
> gcc installed from ports will do the right thing. I haven't checked.
> 
> Oh yeah - if you're running the proprietary NVidia driver, it doesn't
> exist for amd64 on FreeBSD. The open source driver works fine.
> 
> Personally, I think converting is worth it, and would recommend it if
> you can. However, there are things that don't work. The only real way
> to tell if everything you need works is to try and see. If you've got
> critical things that you *have* to have, you might ask here. But the
> answer may not be reliable, because some options that you require may
> not work, like the OpenOffice/Mozilla integration.
> 
> Personally, I bought a second disk identical to my system disk, copied
> the partition table and set up a dual boot system with /home from the
> original system disk on both systems. After I was happy, I moved
> everything to one disk and set them up as a RAID 1.
> 
> While you're doing this, check your motherboard specs, and see if it
> supports dual channel memory access. If it will, buy another stick of
> RAM so that you'll get dual channel mode. That's another relatively
> cheap way to improve system performmance. Since amd64 uses a bit more
> memory, you may want to do this anyway.
> 
> 	<mike